Real-Time Prediction and Management of Intradialytic Hypotension with Machine Learning

Real-Time Prediction and Management of Intradialytic Hypotension with Machine Learning | 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 Article Real-Time Prediction and Management of Intradialytic Hypotension with Machine Learning Yuanhao Wu, Zhu xiaoye, Jianda Lu, Yu Hui, Zhongtian Jiang, Yang Tian, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8786359/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Intradialytic hypotension (IDH) is associated with high mortality and morbidity. This study evaluates the application of machine learning (ML) for IDH diagnosis and management. This study consisted of a prospective real-world study and a pilot randomized controlled trial (RCT). Clinical data from 167 hemodialysis patients (2018–2020) were randomly divided into a training set (75%) and a validation set (25%). ML models (RNN, XGBoost, LightGBM) were assessed under three IDH definitions. The optimal XGBoost model, which utilized a stratified systolic blood pressure (SBP) threshold, achieved a receiver operating characteristic area under the curve (ROC-AUC) of 0.933, demonstrating robust predictive performance. In the RCT, 32 patients were allocated to AI-assisted IDH management or conventional care. Compared to controls, the AI-assisted group had a significantly greater reduction in IDH events (MD − 8.13, 95% CI: − 15.64 to − 0.62, P = 0.034) and a more marked improvement in cumulative SBP decline at IDH onset (MD − 108.69, 95% CI: − 209.83 to − 7.56, P = 0.036). The AI-assisted intervention, based on the XGBoost model predicting IDH risk using a stratified SBP threshold, significantly reduces IDH events, offering a novel strategy for the precise prevention and management of hypotension during dialysis. Clinical Trial registry name and registration number Research on Machine Learning-Based Information Systems for Predicting and Mitigating the Occurence of Intradialytic Hypotension, ChiCTR2000036973 Health sciences/Nephrology/Renal replacement therapy/Haemodialysis Health sciences/Nephrology/Renal replacement therapy Artificial intelligence Hemodialysis Intradialytic hypotension Machine learning AI-assisted intervention Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Intradialytic hypotension (IDH) represents the most prevalent complication in hemodialysis (HD) treatment, and is associated with high mortality 1 . Therefore, timely diagnosis and prompt treatment may be crucial for improving the prognosis of IDH patients. However, there is no consensus on the definition of IDH, which has led to difficulties in data collection and posed a challenge to accurate estimates of the prevalence, risk factors, and preventive measures for IDH 2 , 3 , 4 . Furthermore, in the majority of instances, IDH presents with atypical clinical manifestations, and even experienced nephrologists find it difficult to predict the occurrence risk of IDH. Currently, artificial intelligence (AI) has been employed in predicting IDH; however, machine learning (ML) models developed from electronic health record data often lack robustness in real-world clinical environments 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 . Furthermore, the assessment of artificial intelligence models in the field of nephrology is still mainly based on computer simulations, with a lack of studies on the application of these models in clinical interventions 13 , 14 . Therefore, based on multi-parameter dynamic monitoring and machine learning algorithm screening for the appropriate definition of IDH, the construction of a real-time IDH risk early warning system has significant practical value for enhancing the safety management of hemodialysis. This study evaluated three definitions of IDH using machine learning algorithms. IDH1 was defined as an intradialytic systolic blood pressure (SBP) below 90 mmHg when the predialysis SBP was less than 160 mmHg, or an intradialytic SBP below 100 mmHg when the predialysis SBP was 160 mmHg or higher 1 . IDH2 represented a nadir intradialytic SBP measurement under 90 mmHg 1 . IDH3 constituted a reduction in SBP of at least 20 mmHg from predialysis to intradialytic values 15 , 16 . In this study, we prospectively monitored the dynamic changes in real-time blood pressure during hemodialysis and constructed a predictive model based on machine learning algorithms (Fig. 1 ). Through internal validation and time-split validation, we systematically evaluated the reproducibility and generalization capability of the model. Based on these criteria, we further developed the first dynamic threshold XGBoost based prediction model stratified by pre-dialysis SBP level that can predict IDH prior to an IDH event in real time. Additionally, we assessed the role of the developed machine learning model in the IDH prediction and assisted mamagement via a randomized controlled pilot trial. Results Model Dataset Overview We collected data from 539,044 dialysis sessions involving 215 in-central HD patients from January 1, 2018, to December 31, 2021. Records from cohort 1 (167 patients) included 264,783 records (75%) for the training set and 88,228 records (25%) for the validation set from 2018-01-01 to 2020-12-31(Fig. 2 ). The baseline characteristics of the study population in the validation set, Test-Set 1, Test-Set 2 and Test-Set 3 were similar to those in the training set (Table 1 ). Table 1 Baseline characteristics of the studied population (n = 215) Parameter Training set Validation Set Test Set 1 Test Set 2 Test Set 3 Database of patients 264783 88228 162225 23808 186033 Number of patients 167 167 167 48 215 Male(%) 110(65.8%) 110(65.8%) 110(65.8%) 30(62.5%) 140(65.1%) Diabetes status 16.4% 16.2% 15.5% 24.1% 16.6% Age 60.5 ± 10.9 60.6 ± 10.9 62.2 ± 11.6 63.5 ± 11.1 62.4 ± 11.5 Duration of dialysis (months) 64.8 ± 42 66.0 ± 42 66.0 ± 42 50.4 ± 34.8 63.6 ± 40.8 Predialysis dry weight 64.6 ± 12.5 64.6 ± 12.4 64.8 ± 12.9 64.2 ± 12.1 64.7 ± 12.8 Target ultrafiltration volume 2.70 ± 1.50 2.70 ± 1.00 2.70 ± 1.00 2.30 ± 1.00 2.60 ± 1.00 Interdialytic weight gain 2.20 ± 1.10 2.20 ± 1.10 2.20 ± 1.10 1.90 ± 1.10 2.10 ± 1.10 Predialysis SBP 138.1 ± 24.2 138.2 ± 24.2 137.1 ± 24.1 139.7 ± 24.3 137.4 ± 24.1 Predialysis DBP 77.8 ± 14.7 77.7 ± 14.4 77.8 ± 17.3 79.5 ± 22.9 78.0 ± 18.2 Predialysis heart rate 77.0 ± 12.2 76.9 ± 12.0 76.6 ± 12.2 78.4 ± 12.5 76.8 ± 12.2 Dialysate temperature 36.0 ± 0.5 36.0 ± 0.5 36.0 ± 0.4 36.1 ± 0.4 36.0 ± 0.4 Dialysate conductivity 14.0 ± 0.1 14.0 ± 0.1 14.0 ± 0.1 14.0 ± 0.1 14.0 ± 0.1 Blood flow rate 228.5 ± 78.9 228.0 ± 37.4 234.8 ± 39.5 223.5 ± 34.7 233.3 ± 39.0 Ultrafiltration rate 0.70 ± 0.30 0.70 ± 0.30 0.70 ± 0.30 0.60 ± 0.30 0.70 ± 0.30 Volume of ultrafiltration 1.20 ± 2.40 1.20 ± 10.20 1.20 ± 0.90 1.10 ± 0.80 1.20 ± 0.90 Last SBP 127.1 ± 23.8 126.9 ± 23.8 127.3 ± 23.9 133.2 ± 22.8 128.1 ± 23.7 Last DBP 74.4 ± 14.5 74.2 ± 14.8 74.8 ± 14.9 77.5 ± 16.9 75.2 ± 15.2 Last heart rate 76.7 ± 13.6 76.8 ± 13.7 76.0 ± 13.3 75.6 ± 13.0 75.9 ± 13.3 SBP, systolic blood pressure; DBP, diastolic blood pressure Evaluation of Machine Learning Models and Selection of IDH Definition for Clinical Intervention Based on Predictive Performance and Clinical Feasibility All predictions were evaluated under the best F1 score and proper F1 score for practice. During the period from January 1, 2021 to December 31, 2021 (with a total of 302 dialysis days), the occurrence of IDH in Test-Set 1 under different IDH definitions were discrepant. Among them, the occurrence of IDH3 (51312) was much higher than those of IDH1 (10072) and IDH2(9338) (Table 2 ). In Test-Set 1, the daily number of IDH predicted events was 30, 29, and 163 according to the definitions of IDH1, IDH2, and IDH3 using XGBoost, respectively (LightGBM: 28, 30, 163; RNN: 32, 26, 147, Table 2 ). Table 2 The occurrence of the three IDH definitions in Test Set 1 and the average daily IDH predicted values of each model at the best F1 score. The occurrence of IDH The average daily occurrence of IDH Model Average number of daily IDH predictions IDH1 10072 33.35 XGBoost 30 LightGBM 28 RNN 32 IDH2 9338 30.92 XGBoost 29 LightGBM 30 RNN 26 IDH3 51312 169.91 XGBoost 163 LightGBM 163 RNN 147 IDH, Intradialytic hypotension; XGBoost, Extreme Gradient Boosting; LightGBM, Light Gradient Boosting Machine; RNN, recurrent neural network. We used Test-Set 1, Test-Set 2 and Test-Set 3 to test the performance of the prediction model. Table S1 summarizes the basic parameters of different ML models under the three IDH definitions at the best threshold of the ROC-AUC. Under the definitions of IDH1 and IDH2, the ROC results of the three algorithm models were concentrated in the range of 0.906 to 0.920, with minor variation (less than 0.01). In Test-Set 2, the LightGBM model had the highest ROC-AUC for predicting IDH1 (0.915, 95%CI:0.911–0.918). The Test-Set2/Test-Set1 PR-AUC ratios of IDH1 (RNN: 0.843, XGBoost: 0.892, LightGBM: 0.891) are higher than those of IDH2 (RNN: 0.808, XGBoost: 0.858, LightGBM: 0.859). Similarly, compared with Test-Set 1, the PR-AUC ratios of IDH1 in Test-Set 3 was greater than those of IDH2(Table 3 ). The IDH1 was selected as the IDH definition for clinical intervention based on clinical feasibility and the results of internal validation and time-split validation. Table 3 Changes in the ROC-AUC and PR-AUC of IDH1 and IDH2 in different IDH prediction models. DATA SET ROC-AUC PR-AUC PR-AUC Ratio (Test-Set2/Test-Set1 or Test-Set3 /Test-Set1) RNN IDH1 Test-Set1 0.914 (0.913 to 0.916) 0.517 (0.515 to 0.520) Test-Set2 0.906 (0.902 to 0.910) 0.436 (0.429 to 0.443) 0.843 Test-Set3 0.914 (0.913 to 0.916) 0.511 (0.508 to 0.513) 0.988 IDH2 Test-Set1 0.915 (0.913 to 0.916) 0.506 (0.503 to 0.508) Test-Set2 0.908 (0.904 to 0.912) 0.409 (0.403 to 0.416) 0.808 Test-Set3 0.915 (0.914 to 0.917) 0.498 (0.496 to 0.501) 0.984 XGBoost IDH1 Test-Set1 0.914 (0.912 to 0.915) 0.498 (0.496 to 0.501) Test-Set2 0.914 (0.910 to 0.917) 0.444 (0.438 to 0.450) 0.892 Test-Set3 0.915 (0.913 to 0.916) 0.494 (0.492 to 0.496) 0.992 IDH2 Test-Set1 0.919 (0.918 to 0.921) 0.499 (0.496 to 0.501) Test-Set2 0.918 (0.914 to 0.921) 0.428 (0.422 to 0.434) 0.858 Test-Set3 0.920 (0.919 to 0.921) 0.493 (0.491 to 0.495) 0.988 LightGBM IDH1 Test-Set1 0.913 (0.912 to 0.914) 0.505 (0.503 to 0.508) Test-Set2 0.915 (0.911 to 0.918) 0.450 (0.444 to 0.456) 0.891 Test-Set3 0.914 (0.913 to 0.915) 0.501 (0.498 to 0.503) 0.992 IDH2 Test-Set1 0.917 (0.915 to 0.918) 0.498 (0.496 to 0.501) Test-Set2 0.914 (0.910 to 0.917) 0.428 (0.422 to 0.434) 0.859 Test-Set3 0.917 (0.916 to 0.919) 0.493 (0.490 to 0.495) 0.990 IDH1, intradialytic hypotension defined as SBP < 90 mmHg when predialysis systolic blood pressure < 160 mmHg (or SBP < 100 mmHg when predialysis BP ≥ 160 mmHg); IDH2, intradialytic hypotension defined as nadir systolic BP < 90 mmHg; RNN, recurrent neural network; XGBoost, Extreme Gradient Boosting; LightGBM, light gradient boosting machine; PR-AUC, precision‒recall area under the curve; ROC-AUC, receiver operating characteristic area under the curve. In all IDH definitions, the ROC-AUC curves of the XGBoost, RNN, and LightGBM prediction models all rise steeply to the upper left corner indicates high sensitivity and low false positive rates (Figures S1 , S2, and S3). Model performance was also measured by calibration curves and reached the agreement between predicted and observed outcome (Figures S4). According to the evaluation of F1 score, the best performing predictive model is RNN, followed by XGBoost (Table S1 ). In view of the fact that the RNN model exhibits a high degree of sensitivity to timestamps, following systematic backend testing and a comprehensive assessment of its operational friendliness, tolerance for data missingness, and feasibility in clinical practice, the XGBoost model has been selected for subsequent clinical interventions. Based on the data from January 1, 2018, to July 24, 2023, the XGBoost model was retrained, and the optimal threshold was selected based on the highest F1-score (Table S2 ). Ultimately, the XGBoost model with updated data chose a threshold of 0.3, where the ROC-AUC value was 0.933 and the PR-AUC value was 0.597 (Fig. 4 ). RCT Patient characteristics In total, 32 patients were screened and randomized to the test group or placebo group. The first patient was enrolled on 2023/08/01, and the last patient was enrolled on 2023/08/20 (Fig. 3 ). Table 4 presents the baseline characteristics of patients stratified by randomization. The IDH event incidence rates in the AI-assisted group and the control group one year prior to enrollment were 39.5% (31.3%-77.6%) and 44.4% (36.8%-64.8%), respectively. No statistically significant differences were observed in baseline characteristics between the two groups. Table 4 Baseline Characteristics in AI-assisted and Control Groups AI-assisted group Control group P value Gender (male/female) 10/5 7/8 0.269 Age(year) 65.3 ± 8.5 67.6 ± 10.1 0.498 Duration of dialysis (months) 105.2 ± 68.5 115.4 ± 62.9 0.674 Hypertension(%) 9 (60) 9 (60) 1.000 Cardiovascular disease(%)* 6 (40) 3 (20) 0.427 Hypertension drug use(%) # 7 (46.7) 3 (20) 0.245 Incidence rate of IDH% 39.5 (31.3, 77.6) 44.4 (36.8, 64.8) 0.455 Mean SBP (mmHg) 111.1 ± 16.3 111.3 ± 20.9 0.979 Body mass index (kg/m 2 ) 22.3 ± 3.2 22.3 ± 3.7 0.976 spKt/V 1.60 ± 0.27 1.83 ± 1.16 0.466 Net UFV% 3.84 ± 0.97 3.97 ± 0.98 0.703 IDWG% 4.21 ± 1.06 4.02 ± 1.03 0.628 Serum creatine (mmol/L) 987.1 ± 164.8 1013.4 ± 185.1 0.684 WBC (×10 9 /L) 6.92 ± 1.60 5.92 ± 1.31 0.070 Hemoglobin (g/L) 115.5 ± 10.6 118.9 ± 9.9 0.380 NT-proBNP (pg/mL) 4636 (3468, 6036) 3000 (2318, 7648) 0.534 CRP (mg/L) 4.30 (2.40, 5.00) 5.20 (3.70, 9.10) 0.067 PTH (ng/L) 389.5 (232.1, 469.1) 220.4 (113.1, 385.5) 0.051 BUN (mmol/L) 27.80 ± 5.99 28.64 ± 4.18 0.659 β2-Microglobulin (mg/L) 33.12 ± 3.63 34.10 ± 4.63 0.541 Hematocrit% 35.19 ± 3.15 36.79 ± 3.05 0.168 Albumin (g/L) 39.66 ± 1.99 40.29 ± 3.46 0.545 Phosphate (mmol/L) 1.88 ± 0.32 1.92 ± 0.60 0.823 IDH, intradialytic hypotension; SBP, systolic blood pressure; spKt/V, single pool Kt/V; UFV, ultrafiltration volume; IDWG, interdialytic weight gain; WBC, white blood cells; NT-proBNP, N-terminal pro-B-type natriuretic peptide; CRP, C-reactive protein; PTH, parathyroid hormone; BUN, blood urea nitrogen. *Cardiovascular disease was defined as the presence of either congestive heart failure or atherosclerotic cardiovascular disease, including: previous myocardial infarction, history of coronary artery angioplasty and/or stent placement,angina pectoris, evidence of coronary atherosclerotic disease, stroke, transient ischemic attacks, claudication, or aortic aneurysm. # Medications for treating hypertension should be discontinued on the day of dialysis and only used on non-dialysis days. The Impact of AI-assisted on IDH Occurrence and IDH Incidence The number of IDH occurrences in both groups during the observation period (pre-intervention), intervention period, and evaluation period (post-intervention) did not demonstrate any significant differences (Table 5 ). From pre-intervention to intervention period, in the AI-assisted group, the reduction in the number of IDH occurrences was significantly greater than that observed in the control group (MD -8.13, 95% CI: -15.64 to -0.62, P = 0.034) (Fig. 5 A, Table 5 ). No significant impact of AI assistance on IDH incidence was found (Table S3). Table 5 The Impact of AI-assisted on IDH Occurrence AI-assisted group MD (95% CI) P value Control group MD (95% CI) P value Between-Group difference MD (95% CI) P value Pre-intervention 20.01 (11.47, 28.54) 20.73 (12.19, 29.26) -0.72 (-13.15, 11.71) 0.908 Intervention period 12.34 (3.81, 20.87) 21.19 (12.66, 29.73) -8.86 (-21.29, 3.57) 0.159 Post-intervention 11.14 (2.61, 19.67) 15.66 (7.13, 24.19) -4.52 (-16.95, 7.91) 0.469 Intervention period vs Pre-intervention -7.67 (-12.98, -2.36) 0.005 0.47 (-4.84, 5.78) 0.861 -8.13 (-15.64, -0.62) 0.034 Post-intervention vs Pre-intervention -8.87 (-14.18, -3.56) 0.002 -5.07 (-10.38, 0.24) 0.061 -3.80 (-11.31, 3.71) 0.315 Post-intervention vs Intervention period -1.20 (-6.51, 4.11) 0.653 -5.53 (-10.84, -0.22) 0.041 4.33 (-3.18, 11.84) 0.253 The Impact of AI-assisted on SBP Following the intervention, the AI-assisted group experienced a significantly higher increase in SBP than the control group compared to the pre-intervention period (MD 5.96 mmHg, 95% CI: 0.02 to 11.90, P = 0.049) (Fig. 5 B, Table S4). During pre-intervention, intervention period and post-intervention period, there was no significant difference in the average SBP when IDH occurred in the two groups of patients (Table S5). Throughout the intervention period, including before and after its implementation, no statistically significant difference was observed in the average decrease of SBP between the two groups (Table S6). From pre-intervention to intervention period, the cumulative decrease in SBP within the AI-assisted group was significantly greater than that in the control group (MD -108.69, 95% CI: -209.83 to -7.56, P = 0.036) (Fig. 5 C, Table S7). The Impact of AI-assisted on Patient Clinical Characteristics In comparison to the pre-intervention period, the drop in pre-dialysis BUN in the AI-assisted group was significantly greater than that in the control group (MD -8.09, 95% CI: -11.32 to -4.85, P < 0.001) (Table 6 ). Similarly, in AI-assisted group, compared to the pre-intervention period, the reduction in pre-dialysis serum creatine (Scr) was significantly greater than that in the control group (MD -116.84, 95% CI: -173.30 to -60.37, P < 0.001). In the AI-assisted group, the reduction in white blood cell (WBC) count compared to the pre-intervention period was significantly greater than that observed in the control group (MD -1.15, 95% CI: -1.87 to -0.42, P = 0.003). The variations of other clinical characteristics for the two groups before and after the intervention did not exhibit statistically significant differences (Table S8). 2 , 3 , 4 Table 6 The Impact of AI-assisted on Patient Clinical Characteristics AI-assisted group MD (95% CI) Control group MD (95% CI) Between-Group difference MD (95% CI) P value Pre-intervention 27.95 (25.04, 30.86) 28.50 (25.61, 31.40) -0.55 (-4.78, 3.68) 0.790 BUN (mmol/L) Post-intervention 22.21 (19.25, 25.17) 30.85 (27.90, 33.81) -8.64 (-12.95, -4.33) 0.000 Post-intervention vs Pre-intervention -5.74 (-8.02, -3.45) 2.35 (0.06, 4.64) -8.09 (-11.32, -4.85) < 0.001 Pre-intervention 976.43 (889.15, 1063.71) 1023.07 (936.34, 1109.80) -46.64 (-173.70, 80.42) 0.457 Serum creatine (mmol/L) Post-intervention 833.86 (746.03, 921.70) 997.34 (909.93, 1084.74) -163.48 (-291.44, -35.51) 0.014 Post-intervention vs Pre-intervention -142.57 (-182.50, -102.64) -25.73 (-65.66, 14.20) -116.84 (-173.30, -60.37) < 0.001 Pre-intervention 7.03 (6.15, 7.91) 5.81 (4.94, 6.69) 1.22 (-0.07, 2.50) 0.062 WBC (×10 9 /L) Post-intervention 6.36 (5.48, 7.24) 6.29 (5.40, 7.18) 0.07 (-1.22, 1.36) 0.914 Post-intervention vs Pre-intervention -0.67 (-1.17, -0.17) 0.48 (-0.04, 1.00) -1.15 (-1.87, -0.42) 0.003 BUN, blood urea nitrogen; WBC, white blood cells Discussion In this study, we evaluate and identify the most suitable definition of IDH for AI-based prediction models. Analysis of three definitions—IDH1 (dynamic threshold based on pre-dialysis SBP), IDH2 (fixed threshold), and IDH3 (dynamic change threshold)—revealed distinct characteristics: IDH1 integrates logical symbol reasoning and timestamp attributes; IDH2 represents a single, extreme value that conforms to the underlying pathogenesis and follows binary logical symbol reasoning and decision tree characteristics; while IDH3, which reflects volume response to ultrafiltration during dialysis and reveals the contribution of timestamps to the defined variable, carries a risk of oversensitivity. Previous studies indicate that the frequent occurrence of IDH1 or IDH2 is significantly associated with all-cause mortality 1 , 2 , 3 , 17 , whereas IDH3 show no such correlation 1 . However, the incidence of IDH3 (37.1%-39.2%) was significantly higher than that of IDH2 (2.7%-7.9%) 18, 19 . In this study, the predicted event rate of IDH3 (147–163 events/day) substantially exceeded those of IDH1 (28–32 events/day) and IDH2 (26–30 events/day), confirming that its high sensitivity may lead to overtreatment risks. Previous research on AI-based IDH prediction has primarily utilized retrospective designs, relying on routine data from electronic medical record systems and dialysis equipment 7 , 9 , 10 . While these studies involved multi-center cohorts with large sample sizes, they generally lacked a systematic comparison between internal validation and time-split validation. This limitation introduces uncertainty regarding the models' stability and generalizability in real-world clinical practice 20 . In contrast, our study introduces a novel evaluation framework designed to assess temporal validity through a time-split validation approach. Using a prospectively collected, independent dataset, we systematically evaluated the models' performance in predicting outcomes for new patient populations across different time periods within the same institution. The results show that for both IDH1 and IDH2, the three algorithms maintained excellent ROC-AUC values, ranging from 0.90 to 0.92 across all test sets. The difference between internal validation (Test-Set 1) and time-split validation (Test-Set 2) was less than 0.01, demonstrating strong robustness. Specifically, when using the IDH1 definition, the ratio of PR-AUC between test set 2 and test set 1 was significantly higher than that under the IDH2 definition. Additionally, Test-Set 3 was designed to simulate a dynamic clinical environment with continuous patient enrollment and iterative model updates, closely mimicking real-world hemodialysis center operations. Compared to Test-Set 2, Test-Set 3 not only reflects the natural evolution of patient populations but also supports a more flexible and practical model-updating mechanism. Under this dynamic validation framework, the PR-AUC ratio advantage of the IDH1 definition over Test-Set1 was further amplified, providing practical evidence that the IDH1-based model exhibits enhanced robustness and adaptability in real-world applications. A randomized controlled trial ultimately validated the clinical value of the IDH1 definition. The AI-assisted intervention group showed a significantly greater reduction in the number of IDH occurrences (MD -8.13, P = 0.034) and a larger cumulative decrease in SBP (108.69 mmHg, P = 0.036) compared to the control group. This clinical efficacy was supported by the exceptional discriminative ability of the XGBoost model under the IDH1 definition (ROC-AUC = 0.933). In contrast to existing symptom-dependent definitions (e.g., K/DOQI 2005 15 and KDIGO 2019 21 ), IDH1 enables standardized monitoring through objective parameters, not only improving diagnostic timeliness but also establishing a quantifiable benchmark for quality management in hemodialysis. This study demonstrates that the dynamically threshold-designed IDH1 definition significantly outperforms traditional definitions in both algorithmic robustness and clinical utility, providing a novel paradigm for precise IDH prevention and control. With regard to the predictive performance of IDH, our machine learning models achieved performance levels comparable to those reported in prior studies, while demonstrating distinct advantages. A retrospective study utilizing electronic medical records reported that the RNN model, when defining IDH by Nadir SBP < 90 mmHg, achieved superior performance (ROC-AUC = 0.94) 7 . Consistent with these findings, our RNN model exhibited the highest predictive accuracy (IDH2 Test3 ROC-AUC = 0.915). Nevertheless, considering clinical applicability requirements for missing data handling, we pragmatically selected the secondary-performing XGBoost model for clinical implementation. Another study also defined Nadir90 as the IDH threshold and compared eight machine learning algorithms, ultimately determining XGBoost as the optimal predictor (ROC-AUC = 0.936) 22 . Our research results confirm this conclusion, with the XGBoost model demonstrating robust performance (ROC-AUC = 0.920 for IDH2 Test 3). Notably, a comparative analysis of IDH1 prediction revealed significant research gaps - the existing literature only records one multicenter retrospective study indicating that the performance of logistic regression, random forest, XGBoost, and deep learning models was comparable (ROC-AUC for XGBoost = 0.87) 11 . Our XGBoost model shows significant improvement in this area, achieving higher predictive accuracy (ROC-AUC = 0.915 for IDH1 Test 3). We found that from the pre-intervention to the intervention period, the AI intervention reduced the number of IDH occurrences. This observation may be attributed to the timely detection and management of IDH, which allowed for the rapid infusion of physiological saline or dialysate, improving vascular replenishment and reducing the temporary osmotic gradient between intracellular and extracellular compartments during dialysis 23 , 24 , 25 . Additionally, this may potentially be associated with the elongation of the corresponding dialysis period. Previous studies have revealed that increasing the treatment duration could be beneficial for patients with hemodynamic or cardiovascular instability, significantly lowering the incidence of IDH 26 , 27 . Furthermore, the definition of IDH cannot reflect the cumulative damage to various organs caused by repeated IDH occurrences during each dialysis session 28 , 29 , while the cumulative decrease in SBP may better reflect the impact of IDH recurrence on patients. The cumulative decrease in SBP refers to the cumulative amount of reduction in SBP from the IDH threshold to the occurrence of IDH. The cumulative decline of SBP is employed to assess the degree of impairment caused by recurrent IDH to the patient. Our study showed that the AI-assisted group had a significantly better SBP cumulative decrease than the control group from the pre-intervention to the intervention period. This suggests that AI-assisted intervention may reduce the cumulative damage to organs caused by IDH recurrence in patients. In this study, although the number of IDH occurrences in the AI-assisted group was significantly lower than that in the control group (p = 0.034), the difference in the IDH incidence between the two groups did not reach statistical significance (p > 0.05). This discrepancy may be attributed to insufficient statistical power due to the limited sample size. Future research should prioritize multi-center, large-scale randomized controlled trials with expanded sample sizes to enhance statistical power, thereby validating the clinical utility of the proposed model in reducing IDH incidence. Our research design, data density and data quality offer considerable advantages. First, we prespecified the study protocol and prospectively collected the data for the study design. Currently, all ML model studies predicting IDH are based on retrospective data 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 . Additionally, as most ML models use data from multiple sources, including machine data, electronic health records, and other sources, extracting and merging such multimodal data can be challenging 30 . The performance and generalizability of the model may be affected by the poor quality of some of the data or incomplete data, leading to information bias. In contrast, a prospective research design can overcome this limitation. This model training method is different from those used in previous ML studies 5 , 6 , 7 , 8 , 9 . Furthermore, our server collects hemodialysis data every minute and prospectively collects BP measurements every 30 minutes, thereby increasing the data density per unit time, resulting in only a 30-minute interval between the training sets of our models. Compared with traditional hourly data, collecting BP data every 30 minutes also increases the data volume by 80%, providing a more accurate time window for preventive intervention and greater potential clinical value 7 , 31 . In terms of data quality, we have strict criteria for patient inclusion, requiring a minimum of one year of outpatient dialysis at the study hospital. In our study, the quality of patient data was controlled by quality control specialists, and patients with poor data quality, such as poor compliance and a high degree of missing data, were excluded. High-quality data can be used as training data to enhance the model's ability, improve its effectiveness, reliability, and stability, reduce the risk of overfitting, and enhance its robustness and generalizability. Therefore, AI-assisted intervention, based on our model, can still effectively reduce the occurrences of IDH in this complex, nonlinear, and heterogeneous pathophysiological process of IDH. Currently, the means of achieving timely diagnosis and intervention for IDH rely on dialysis equipment with online blood volume monitoring functions, data feedback mechanisms, and support from healthcare professionals 21 , 28 . However, such hardware devices are typically associated with substantial costs. Our model can be implemented in dialysis facilities with lower hardware requirements, enabling them to use ML models to optimize computational algorithms, relying only on basic dialysis equipment to achieve real-time blood pressure monitoring and IDH occurrence prediction alerts. This study has several limitations. Primarily, there might be a certain amount of missing data in the dataset. Additionally, the sample size of this study is relatively limited and was conducted in a single center, which may restrict the applicability of the results in other healthcare settings with distinct patient populations and resources. However, owing to the methodological design and inherent characteristics of the machine learning dataset, the proposed model exhibits strong scalability, effectively mitigating these limitations. It can be compatible with data from other dialysis centers and generate models that simulate the real local environment, thus providing guidance for early diagnosis and intervention by local IDH. In conclusion, our model can be used to predict real-time risk of IDH defined by different nadir SBP based on pre-HD SBP stratification, and AI assisted intervention based on this model can significantly reduce the occurrence of IDH. Methods Study Design This study is a combination of prospective real-world research and a randomized controlled trial (chictr.org.cn: ChiCTR2000036973). All participants provided written informed consent before the study began. We collected demographic data and hemodialysis records for each participant. The study protocol has been approved by the ethics review board of Huashan Hospital (affiliated to Fudan University) (IRB number: KY2021-609). Our study consists of two parts: 1) a prospective real-world study, including data collection, models establishment and systematic internal and time-split validation of the models (Fig. 1 , 2 ); 2) a randomized controlled study in which the model is applied for AI intervention (Fig. 3 ). In the data collection, model establishment, and internal and time-split validation of the models section, we enrolled adult patients with ESRD who received regular hemodialysis at the Hemodialysis Center of Huashan Hospital Affiliated to Fudan University, Baoshan Branch (January 1, 2018 - December 31, 2021). Eligible ESRD patients were at least 18 years old and had completed one year of uninterrupted outpatient dialysis at this hospital. Exclusion criteria are detailed in Supplemental Methods. In the prospective cohort (Cohort 1) enrolled between January 1, 2018 and December 31, 2020, blood pressure was monitored using a standard upper-arm cuff for each patient during every hemodialysis session. Measurements were recorded at 30-minute intervals, ensuring the acquisition of at least nine blood pressure readings per dialysis session (Fig. 2 ) 7 , 31 , 32 . Data from Cohort 1 included demographic, clinical and laboratory data obtained from electronic health records and dialysis machine data are detailed in Supplemental Methods. Each record was treated as a basic unit (Fig. 1 ). Seventy-five percent of the data were randomly assigned to the training set, while the remaining 25% was used for validation set. We collected dialysis data of cohort 1 (original patients) during the treatment process from January 1, 2021, to December 31, 2021, for internal validation (Test-Set 1), including real-time monitoring blood pressure and prediction IDH in the new follow-up data. At the same time, we collected a new data Test-Set (Test-Set 2) prospectively for time-split validation of patients starting from January 1, 2021, to December 31, 2021 (cohort 2, new enrolled patients). The data of Test-Set 1 and Test-Set 2 were combined to form Test-Set 3, which maybe in accordance with clinical practice. In AI intervention part, we recruited adult HD patients (2022/07-2023/07) receiving routine hemodialysis at the same Hospital and in the initial model training data set. Participants should be able to tolerate high-flux hemodialysis (HFHD) and Fresenius FX80 dialyzers (Fresenius Medical Care, Bad Homburg, Germany). The incidence of IDH in participants must be at least 30% within one year prior to enrollment. Other inclusion and exclusion criteria are described in Supplemental Methods. Definition and Classification of IDH Definition 1 (IDH1): intradialysis SBP < 90 mmHg when the predialysis SBP < 160 mmHg or intradialysis SBP < 100 mmHg when the predialysis SBP ≥ 160 mmHg 1 . Definition 2 (IDH2): A nadir intradialysis SBP < 90 mmHg 1 . Definition 3 (IDH3): A reduction in SBP of ≥ 20 mmHg from predialysis to intradialysis levels 15 , 16 . We regarded each IDH criterion as an independent binary outcome (0 or 1). The overall incidence of IDH was the proportion of hemodialysis sessions with IDH. Construction and Implementation of a Multi-Source Heterogeneous Hemodialysis Database for Intelligent Prediction of IDH This study developed a machine learning-oriented hemodialysis database based on the commercially mature Hemodialysis Information System from GCSOFT Company Limited. The database integrates multi-source heterogeneous data, including patient diagnosis and treatment information related to hemodialysis, hemodialysis machine operational data, and laboratory test results interfaced with the hospital's Hospital Information System (HIS). Building upon this foundation, this study prospectively and innovatively established a structured data acquisition module. Objective data primarily include pre-set or automatically collectable technical parameters and dynamic treatment process data, which are automatically uploaded to the server in real-time via system interfaces. Specifically, this includes: hemodialysis mode settings, ultrafiltration rates at various time points, periods of ultrafiltration pause, dialysate flow rate, conductivity, temperature, ultrafiltration mode, blood flow rate, isolated ultrafiltration periods, blood pressure and heart rate measured every 30 minutes using machine-coupled cuffs, machine-based fluid replenishment records, and records of non-machine-based fluid replenishment orders. All adjustments to medical orders or machine parameters are logged in real-time within the information system and synchronously uploaded to the server, ensuring data integrity and timeliness. Machine Learning Prediction Models and Evaluation Metrics We explored three different ML prediction models: the recurrent neural network (RNN) model 18 ; Extreme Gradient Boosting (XGBoost) 33 and Light Gradient Boosting Machine (LightGBM). This study employed a comprehensive set of metrics to evaluate model performance, including the F1 score, precision‒recall area under the curve (PR-AUC), and receiver operating characteristic area under the curve (ROC-AUC), among others. In our study, model calibration was performed using the Platt scaling method based on test set data. The Supplemental Methods provides detailed explanations of all evaluation metrics and the research framework for ML prediction models. Randomization and Intervention Through a randomly generated sequence by the computer, patients were randomly assigned in a 1:1 ratio into two groups: one group received manual intervention after the ML model diagnosed IDH (AI-assisted group), and the other group received standard operation (the control group). The random sequence was generated by an independent database programmer prior to the trial to ensure blinding until the day before the trial. The first-line doctors were not involved in this study and had no knowledge of the random allocation. After both groups of patients were enrolled, they would go through a 4-week observation/washout period (pre-intervention), a 4-week intervention period, and a 4-week evaluation period (post-intervention). During the entire 12-week trial phase, the AI system would monitor in real-time the blood pressure and dialysis parameters of all dialysis patients, and record the occurrence of IDH. In this study, ultra-pure dialysate was provided by the central dialysate delivery system (CDDS), and hemodialysis treatment was performed using GC-110N fully automated dialysis machines (JMS Co. Ltd., Tokyo, Japan). Hemodialysis mode was all HFHD. During the observation/washout period, all patients who originally received hemodialysis/hemofiltration/hemoperfusion treatment were adjusted to receive three sessions of HFHD per week. During the intervention period, when the AI system predicted an IDH event in the AI-assisted group at the next time point, a pop-up window for intervention prompting would appear on the computer in the nursing station (Fig. 2 ). The nurse instructed the patient to restrict their diet, offered dietary education, adjusted the temperature of the dialysis fluid to 35°C, and maintained the patient in Trendelenburg position as soon as possible. The nurse is required to administer 100 milliliters of ultrapure dialysate or normal saline intravenously. The duration of each dialysis session should be dynamically adjusted based on the patient's ultrafiltration rate. If necessary, the ultrafiltration rate can be moderately increased within the clinically acceptable range. In the control group, the AI system would predict IDH events, which were only recorded on the computer terminal without a window for intervention popping up. The intervention in the control group was based on clinical-driven operation, i.e., following the usual method when patients reported intolerance, described in Supplemental Methods. During the evaluation period, the AI system would monitor blood pressure in real-time for both groups of patients and predict IDH events, only recorded on the computer terminal. Prespecified Endpoints Primary Endpoint: The number of IDH occurrences during hemodialysis sessions. Secondary Endpoints: Included the incidence of IDH during hemodialysis, mean SBP during dialysis, mean SBP at IDH onset, mean reduction in SBP from the IDH threshold to the occurrence of IDH, and cumulative reduction in SBP from the IDH threshold to IDH occurrence. Exploratory Endpoint: Differences in clinical characteristics between the two groups post-intervention. Data characteristics of randomized controlled trial Demographic, dialysis-related, and laboratory data were collected from enrolled patients during the study period, including gender, age, years on dialysis, the history of hypertension and cardiovascular disease, drugs used anti-hypertension, body mass index (BMI), Kt/V value, average weight gain during the dialysis interval relative to baseline, average net ultrafiltration relative to baseline, and creatinine level before dialysis, white blood cell (WBC) count, hemoglobin concentration, hematocrit, high-sensitivity C-reactive protein level, albumin concentration, electrolyte parameters (phosphorus), intact parathyroid hormone, and N-terminal pro-brain natriuretic peptide (NT-proBNP). Statistical analysis This study was designed as an Early Feasibility Study (EFS). The sample size of 32 patients was determined based on feasibility considerations to evaluate initial safety and device performance, in accordance with FDA guidelines for EFS, rather than a formal power calculation for confirmatory efficacy 34 . Baseline characteristics of the experimental and control groups were summarized using descriptive statistics. Continuous variables were presented as mean ± standard deviation or median with interquartile range for skewed distributions. Group comparisons at baseline were performed using the Student’s t-test or Wilcoxon rank-sum test, as appropriate. For the efficacy analysis, a linear mixed model for repeated measures was employed to compare the groups. The model included the baseline PTH level, time, treatment group, and the group-by-time interaction as fixed effects. A logarithmic transformation was applied to variables with skewed distributions prior to model fitting to satisfy normality assumptions. The treatment effect is expressed as the least-squares mean difference (LSMD) with its 95% confidence interval (CI). All analyses were performed using SAS 9.4.. Declarations Disclosures There are no conflicts of interest to disclose. Funding This work was supported by the Scientific Research Project of Shanghai Municipal Health Commission, No. 201940271. Author Contributions YW, XZhu, JL, and CH conceived and designed the study. XZhou designed the clinical trial section. YW, XZhu, and JL provided data interpretation. JL, YH and XZhou conducted the data analysis. YW,XZhu and JL drafted and revised the article. YW, XZhu, JL, LY, XL, YR and JX were responsible for data collection. YW, LY, XL and YR performed patient recruitment. YW and JL configured and adjusted the logic and parameters of the machine learning model. YH and JL are responsible for machine learning coding and data maintenance. YH and JL tested the machine learning model. YT and ZJ are responsible for utilizing the program front-end and data maintenance. YW, XZhu and JL contributed equally to this work. All the authors reviewed and approved the manuscript for submission. Data Sharing Statement The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Code availability https://github.com/WuYuanhaoSlug/IDH_Prediction References Flythe JE, Xue H, Lynch KE, Curhan GC, Brunelli SM (2015) Association of mortality risk with various definitions of intradialytic hypotension. J Am Soc Nephrol 26:724–734 Assimon MM, Flythe JE (2017) Definitions of intradialytic hypotension. Semin Dial 30:464–472 Kuipers J et al (2019) The Prevalence of Intradialytic Hypotension in Patients on Conventional Hemodialysis: A Systematic Review with Meta-Analysis. Am J Nephrol 49:497–506 Sands JJ et al (2014) Intradialytic hypotension: frequency, sources of variation and correlation with clinical outcome. Hemodial Int 18:415–422 Yang X et al (2022) Boosted machine learning model for predicting intradialytic hypotension using serum biomarkers of nutrition. Comput Biol Med 147:105752 Yang X et al (2022) An optimized machine learning framework for predicting intradialytic hypotension using indexes of chronic kidney disease-mineral and bone disorders. Comput Biol Med 145:105510 Lee H et al (2021) Deep Learning Model for Real-Time Prediction of Intradialytic Hypotension. Clin J Am Soc Nephrol 16:396–406 Liu YS et al (2021) Machine Learning Analysis of Time-Dependent Features for Predicting Adverse Events During Hemodialysis Therapy: Model Development and Validation Study. J Med Internet Res 23:e27098 Dong J et al (2023) Machine learning-based intradialytic hypotension prediction of patients undergoing hemodialysis: A multicenter retrospective study. Comput Methods Programs Biomed 240:107698 Zhang H et al (2023) Real-time prediction of intradialytic hypotension using machine learning and cloud computing infrastructure. Nephrol Dial Transpl 38:1761–1769 Lee H et al (2023) Prediction of intradialytic hypotension using pre-dialysis features-a deep learning-based artificial intelligence model. Nephrol Dial Transpl 38:2310–2320 Huang JC et al (2020) Predictive modeling of blood pressure during hemodialysis: a comparison of linear model, random forest, support vector regression, XGBoost, LASSO regression and ensemble method. Comput Methods Programs Biomed 195:105536 Escobar GJ, Liu VX, Schuler A, Lawson B, Greene JD, Kipnis P (2020) Automated Identification of Adults at Risk for In-Hospital Clinical Deterioration. N Engl J Med 383:1951–1960 Loftus TJ et al (2022) Artificial intelligence-enabled decision support in nephrology. Nat Rev Nephrol 18:452–465 K/DOQI clinical (2005) practice guidelines for cardiovascular disease in dialysis patients. Am J Kidney Dis 45:S1–153 Kooman J et al (2007) EBPG guideline on haemodynamic instability. Nephrol Dial Transpl 22(2):ii22–44 Wu Y, Lu J, Wang T, Zhu X, Xue J, You L (2024) Association of frequent intradialytic hypotension with the clinical outcomes of patients on hemodialysis: a prospective cohort study. Ren Fail 46:2296612 Rank N et al (2020) Deep-learning-based real-time prediction of acute kidney injury outperforms human predictive performance. NPJ Digit Med 3:139 Kim HW et al (2022) Deep Learning Model for Predicting Intradialytic Hypotension Without Privacy Infringement: A Retrospective Two-Center Study. Front Med (Lausanne) 9:878858 de Hond AAH, Shah VB, Kant IMJ, Van Calster B, Steyerberg EW, Hernandez-Boussard T (2023) Perspectives on validation of clinical predictive algorithms. NPJ Digit Med 6:86 Flythe JE et al (2020) Blood pressure and volume management in dialysis: conclusions from a Kidney Disease: Improving Global Outcomes (KDIGO) Controversies Conference. Kidney Int 97, 861–876 Yang IN et al (2024) Personalized prediction of intradialytic hypotension in clinical practice: Development and evaluation of a novel AI dashboard incorporating risk factors from previous and current dialysis sessions. Int J Med Inf 190:105538 Arieff AI, Massry SG, Barrientos A, Kleeman CR (1973) Brain water and electrolyte metabolism in uremia: effects of slow and rapid hemodialysis. Kidney Int 4:177–187 Kennedy AC, Linton AL, Eaton JC (1962) Urea levels in cerebrospinal fluid after haemodialysis. Lancet 1:410–411 Silver SM, Sterns RH, Halperin ML (1996) Brain swelling after dialysis: old urea or new osmoles? Am J Kidney Dis 28:1–13 Brunet P, Saingra Y, Leonetti F, Vacher-Coponat H, Ramananarivo P, Berland Y (1996) Tolerance of haemodialysis: a randomized cross-over trial of 5-h versus 4-h treatment time. Nephrol Dial Transpl 11(Suppl 8):46–51 Locatelli F, Manzoni C (1999) Duration of dialysis sessions–was Hegel right? Nephrol Dial Transpl 14:560–563 Reeves PB, Mc Causland FR, Mechanisms (2018) Clinical Implications, and Treatment of Intradialytic Hypotension. Clin J Am Soc Nephrol 13:1297–1303 LSQ NN (2023) A comparative study of the definitions of intradialytic hypotension correlated with increased mortality to identify universal predictors. Int J Med Inf 173:104975 Kotanko P, Zhang H, Wang Y (2023) Artificial Intelligence and Machine Learning in Dialysis: Ready for Prime Time? Clin J Am Soc Nephrol 18:803–805 Lin CJ, Chen YY, Pan CF, Wu V, Wu CJ (2019) Dataset supporting blood pressure prediction for the management of chronic hemodialysis. Sci Data 6:313 Sinha AD, Agarwal R (2009) Peridialytic, intradialytic, and interdialytic blood pressure measurement in hemodialysis patients. Am J Kidney Dis 54:788–791 Chen T, Guestrin C, XGBoost: (2016) A Scalable Tree Boosting System. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining ). Association for Computing Machinery FDA. Investigational Device Exemptions (IDEs) for Early Feasibility Medical Device Clinical Studies, Including Certain First in Human (FIH) Studies.) (2013) Additional Declarations There is NO Competing Interest. Supplementary Files SupplementaryAppendix.docx Supplementary appendix Cite Share Download PDF Status: Under Review 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-8786359","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":605348087,"identity":"5b709b81-144b-4b37-8106-11a58afdacc2","order_by":0,"name":"Yuanhao Wu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAArElEQVRIiWNgGAWjYDACdgglR4IWZghlTLqWxAaidcg38xi/YGw7nL62/+wBhh8V2whrMTjMY2YB1JK77UZeAmPPmdtEaGHmMTNg3AbSwmPAzNhGhBagw8Ba0s3OnyFSC8NhHuMHQC0JZgdyiNRicJitjIHxX7rhths5BgeJ8ot8e/PmDwxnrOWBDjN88KOCGIcxMLBJ/4GyDhClHgiYPxCrchSMglEwCkYoAADFlzg5qKp8xwAAAABJRU5ErkJggg==","orcid":"","institution":"Fudan University Huashan Hospital","correspondingAuthor":true,"prefix":"","firstName":"Yuanhao","middleName":"","lastName":"Wu","suffix":""},{"id":605348088,"identity":"b7035ee0-5bfc-4275-9f56-62a9e36e3d3f","order_by":1,"name":"Zhu xiaoye","email":"","orcid":"https://orcid.org/0000-0002-9855-9122","institution":"Fudan University Huashan Hospital","correspondingAuthor":false,"prefix":"","firstName":"Zhu","middleName":"","lastName":"xiaoye","suffix":""},{"id":605348089,"identity":"07188c2c-0bdd-40a3-8cd1-fa4668451190","order_by":2,"name":"Jianda Lu","email":"","orcid":"","institution":"fudan university huashan hospital","correspondingAuthor":false,"prefix":"","firstName":"Jianda","middleName":"","lastName":"Lu","suffix":""},{"id":605348090,"identity":"352b2192-209f-4e27-8e44-72c72eae5558","order_by":3,"name":"Yu Hui","email":"","orcid":"","institution":"GCSOFT CO.,LTD","correspondingAuthor":false,"prefix":"","firstName":"Yu","middleName":"","lastName":"Hui","suffix":""},{"id":605348091,"identity":"4db87346-98e8-4a89-9a4d-6c42d0f38ff4","order_by":4,"name":"Zhongtian Jiang","email":"","orcid":"","institution":"GCSOFT CO.,LTD","correspondingAuthor":false,"prefix":"","firstName":"Zhongtian","middleName":"","lastName":"Jiang","suffix":""},{"id":605348092,"identity":"ba7a77eb-21c4-4f6d-92d2-264e47df5be8","order_by":5,"name":"Yang Tian","email":"","orcid":"","institution":"GCSOFT CO.,LTD","correspondingAuthor":false,"prefix":"","firstName":"Yang","middleName":"","lastName":"Tian","suffix":""},{"id":605348093,"identity":"2713a983-dfb6-4004-b9c1-1de09a6e972e","order_by":6,"name":"Xiaowan Liang","email":"","orcid":"","institution":"fudan university huashan hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiaowan","middleName":"","lastName":"Liang","suffix":""},{"id":605348094,"identity":"55474076-33e9-42b3-972a-6b3bc45e930f","order_by":7,"name":"Yuan Ren","email":"","orcid":"","institution":"fudan university huashan hospital","correspondingAuthor":false,"prefix":"","firstName":"Yuan","middleName":"","lastName":"Ren","suffix":""},{"id":605348095,"identity":"b0a1274b-ce00-4601-8ef3-23e80ca66321","order_by":8,"name":"Xu Zhou","email":"","orcid":"","institution":"Jiangxi University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Xu","middleName":"","lastName":"Zhou","suffix":""},{"id":605348096,"identity":"01cacae5-6e47-46cb-880f-7134bd104af7","order_by":9,"name":"Li You","email":"","orcid":"","institution":"Huashan North Hospital Baoshan Branch, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"You","suffix":""},{"id":605348097,"identity":"bec13940-8797-4f2c-95ca-ff00561de191","order_by":10,"name":"Jun Xue","email":"","orcid":"","institution":"Huashan Hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Jun","middleName":"","lastName":"Xue","suffix":""},{"id":605348098,"identity":"374afc32-62e0-47d8-b1c1-77a2d792fcca","order_by":11,"name":"Chuanming Hao","email":"","orcid":"","institution":"Division of Nephrology, Huashan Hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Chuanming","middleName":"","lastName":"Hao","suffix":""}],"badges":[],"createdAt":"2026-02-04 12:13:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8786359/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8786359/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104874525,"identity":"14215ad9-0f4a-4486-96d0-59f4b3e22010","added_by":"auto","created_at":"2026-03-18 08:31:48","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":370858,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSchematic diagram of the ML model prediction and feedback\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBlood pressure was automatically monitored during all hemodialysis treatments, with systolic blood pressure (SBP) and diastolic blood pressure (DBP) data collected at 30-minute intervals. Machine learning model of RNN, XGBoost, and LightGBM to predict the IDH occurrence within the subsequent 30-minute period. At the application level, the system achieves intelligent integration of hemodialysis: through a wired local area network architecture, multimodal physiological and management data collected by self-service weighing devices, non-invasive blood pressure monitors, dialysis equipment, and identity recognition scanners are transmitted in real-time to a central data acquisition server. The artificial intelligence server performs feature extraction and risk assessment on the preprocessed data to generate IDH early warning signals. The application service layer then dynamically delivers structured prediction results to clinical terminal systems (such as nursing stations and physician consoles) through an adaptive scheduling mechanism, realizing closed-loop management from data acquisition to clinical response.\u003c/p\u003e\n\u003cp\u003eRNN, recurrent neural network; XGBoost, Extreme Gradient Boosting; LightGBM, Light Gradient Boosting Machine; SBP, systolic blood pressure; MAP, mean arterial pressure; IDH, intradialytic hypotension.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8786359/v1/205e82c1036ea3e9692f4a44.png"},{"id":105034272,"identity":"33767a1e-42eb-419b-beff-7faa3db2186f","added_by":"auto","created_at":"2026-03-20 07:23:00","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":446498,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eResearch workflow of the ML model development and validation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eXGBoost, Extreme Gradient Boosting; RNN, recurrent neural network; LightGBM, light gradient boosting machine; PR-AUC, precision‒recall area under the curve; ROC-AUC, receiver operating characteristic area under the curve; IDH, intradialytic hypotension.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8786359/v1/eedac99d22e69b30ccc56f93.png"},{"id":105034171,"identity":"d2157c75-088c-4406-b8a4-a1b4ebad34cf","added_by":"auto","created_at":"2026-03-20 07:22:48","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":601902,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eResearch workflow of the randomized controlled study.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIDH, intradialytic hypotension; HFHD, high-flux hemodialysis; ML, machine learning.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8786359/v1/74e1b83f796e9c73d2ddb770.png"},{"id":104874520,"identity":"f7674743-7c0b-4181-896b-e1fa72903767","added_by":"auto","created_at":"2026-03-18 08:31:48","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":104567,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eReceiver operating characteristic (A) and precision-recall (B) curves for XGBoost prediction of IDH1.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eROC-AUC, receiver operating characteristic area under the curve; PR-AUC, precision‒recall area under the curve.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8786359/v1/f78b8ca242bbe63fbc9a6fbf.png"},{"id":104874522,"identity":"fcfce25f-a08c-4273-a69f-c01d888d1e55","added_by":"auto","created_at":"2026-03-18 08:31:48","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":543816,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe occurrence of IDH, mean systolic blood pressure, cumulative decrease in systolic blood pressure in the two groups\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe occurrence of IDH in the two groups. IDH, intradialytic hypotension, *From pre-intervention to intervention period, the difference of IDH occurrence between two groups was significantly (MD -8.13, 95% CI: -15.64 to -0.62, P=0.034). (B) The systolic blood pressure within the two groups. SBP, systolic blood pressure, *From post-intervention to intervention period, the AI-assisted group experienced a significantly higher increase in mean SBP than the control group (MD 5.96 mmHg, 95% CI: 0.02 to 11.90, P =0.049). (C) The cumulative decrease in systolic blood pressure within two groups.The cumulative decrease in SBP refers to the cumulative amount of reduction in systolic blood pressure from the IDH threshold to the occurrence of IDH. The cumulative decline of SBP is employed to assess the degree of impairment caused by recurrent IDH to the patient. SBP, systolic blood pressure, *From post-intervention to intervention period, the cumulative decrease in SBP within the AI-assisted group was significantly greater than that in the control group (MD -108.69, 95% CI: -209.83 to -7.56, P=0.036).\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8786359/v1/cd1b54a164e6710cfc4f2986.png"},{"id":105562947,"identity":"b1626a30-d7a9-4f1e-bcff-e7720f4dfbf3","added_by":"auto","created_at":"2026-03-27 12:45:21","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3509643,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8786359/v1/5e273128-4893-44b4-a08f-2ee45eb2dafc.pdf"},{"id":104874524,"identity":"95bec44e-20c3-4bb3-a599-05c2be4fe733","added_by":"auto","created_at":"2026-03-18 08:31:48","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1316490,"visible":true,"origin":"","legend":"Supplementary appendix","description":"","filename":"SupplementaryAppendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-8786359/v1/7f1bfc667e5cfaa4401b79c2.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Real-Time Prediction and Management of Intradialytic Hypotension with Machine Learning","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIntradialytic hypotension (IDH) represents the most prevalent complication in hemodialysis (HD) treatment, and is associated with high mortality \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Therefore, timely diagnosis and prompt treatment may be crucial for improving the prognosis of IDH patients.\u003c/p\u003e \u003cp\u003eHowever, there is no consensus on the definition of IDH, which has led to difficulties in data collection and posed a challenge to accurate estimates of the prevalence, risk factors, and preventive measures for IDH \u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Furthermore, in the majority of instances, IDH presents with atypical clinical manifestations, and even experienced nephrologists find it difficult to predict the occurrence risk of IDH.\u003c/p\u003e \u003cp\u003eCurrently, artificial intelligence (AI) has been employed in predicting IDH; however, machine learning (ML) models developed from electronic health record data often lack robustness in real-world clinical environments \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Furthermore, the assessment of artificial intelligence models in the field of nephrology is still mainly based on computer simulations, with a lack of studies on the application of these models in clinical interventions \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Therefore, based on multi-parameter dynamic monitoring and machine learning algorithm screening for the appropriate definition of IDH, the construction of a real-time IDH risk early warning system has significant practical value for enhancing the safety management of hemodialysis.\u003c/p\u003e \u003cp\u003eThis study evaluated three definitions of IDH using machine learning algorithms. IDH1 was defined as an intradialytic systolic blood pressure (SBP) below 90 mmHg when the predialysis SBP was less than 160 mmHg, or an intradialytic SBP below 100 mmHg when the predialysis SBP was 160 mmHg or higher\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. IDH2 represented a nadir intradialytic SBP measurement under 90 mmHg\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. IDH3 constituted a reduction in SBP of at least 20 mmHg from predialysis to intradialytic values\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn this study, we prospectively monitored the dynamic changes in real-time blood pressure during hemodialysis and constructed a predictive model based on machine learning algorithms (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Through internal validation and time-split validation, we systematically evaluated the reproducibility and generalization capability of the model. Based on these criteria, we further developed the first dynamic threshold XGBoost based prediction model stratified by pre-dialysis SBP level that can predict IDH prior to an IDH event in real time. Additionally, we assessed the role of the developed machine learning model in the IDH prediction and assisted mamagement via a randomized controlled pilot trial.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eModel Dataset Overview\u003c/h2\u003e \u003cp\u003eWe collected data from 539,044 dialysis sessions involving 215 in-central HD patients from January 1, 2018, to December 31, 2021. Records from cohort 1 (167 patients) included 264,783 records (75%) for the training set and 88,228 records (25%) for the validation set from 2018-01-01 to 2020-12-31(Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The baseline characteristics of the study population in the validation set, Test-Set 1, Test-Set 2 and Test-Set 3 were similar to those in the training set (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline characteristics of the studied population (n\u0026thinsp;=\u0026thinsp;215)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameter\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eValidation Set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTest Set 1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTest Set 2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTest Set 3\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDatabase of patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e264783\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e88228\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e162225\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e23808\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e186033\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber of patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e215\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e110(65.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e110(65.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e110(65.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e30(62.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e140(65.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetes status\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e16.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e24.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e16.6%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e60.5\u0026thinsp;\u0026plusmn;\u0026thinsp;10.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60.6\u0026thinsp;\u0026plusmn;\u0026thinsp;10.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e62.2\u0026thinsp;\u0026plusmn;\u0026thinsp;11.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e63.5\u0026thinsp;\u0026plusmn;\u0026thinsp;11.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e62.4\u0026thinsp;\u0026plusmn;\u0026thinsp;11.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDuration of dialysis (months)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e64.8\u0026thinsp;\u0026plusmn;\u0026thinsp;42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e66.0\u0026thinsp;\u0026plusmn;\u0026thinsp;42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e66.0\u0026thinsp;\u0026plusmn;\u0026thinsp;42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e50.4\u0026thinsp;\u0026plusmn;\u0026thinsp;34.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e63.6\u0026thinsp;\u0026plusmn;\u0026thinsp;40.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredialysis dry weight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e64.6\u0026thinsp;\u0026plusmn;\u0026thinsp;12.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e64.6\u0026thinsp;\u0026plusmn;\u0026thinsp;12.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e64.8\u0026thinsp;\u0026plusmn;\u0026thinsp;12.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e64.2\u0026thinsp;\u0026plusmn;\u0026thinsp;12.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e64.7\u0026thinsp;\u0026plusmn;\u0026thinsp;12.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTarget ultrafiltration volume\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.70\u0026thinsp;\u0026plusmn;\u0026thinsp;1.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.70\u0026thinsp;\u0026plusmn;\u0026thinsp;1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.70\u0026thinsp;\u0026plusmn;\u0026thinsp;1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.30\u0026thinsp;\u0026plusmn;\u0026thinsp;1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.60\u0026thinsp;\u0026plusmn;\u0026thinsp;1.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInterdialytic weight gain\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.20\u0026thinsp;\u0026plusmn;\u0026thinsp;1.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.20\u0026thinsp;\u0026plusmn;\u0026thinsp;1.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.20\u0026thinsp;\u0026plusmn;\u0026thinsp;1.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.90\u0026thinsp;\u0026plusmn;\u0026thinsp;1.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.10\u0026thinsp;\u0026plusmn;\u0026thinsp;1.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredialysis SBP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e138.1\u0026thinsp;\u0026plusmn;\u0026thinsp;24.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e138.2\u0026thinsp;\u0026plusmn;\u0026thinsp;24.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e137.1\u0026thinsp;\u0026plusmn;\u0026thinsp;24.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e139.7\u0026thinsp;\u0026plusmn;\u0026thinsp;24.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e137.4\u0026thinsp;\u0026plusmn;\u0026thinsp;24.1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredialysis DBP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e77.8\u0026thinsp;\u0026plusmn;\u0026thinsp;14.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e77.7\u0026thinsp;\u0026plusmn;\u0026thinsp;14.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e77.8\u0026thinsp;\u0026plusmn;\u0026thinsp;17.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e79.5\u0026thinsp;\u0026plusmn;\u0026thinsp;22.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e78.0\u0026thinsp;\u0026plusmn;\u0026thinsp;18.2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredialysis heart rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e77.0\u0026thinsp;\u0026plusmn;\u0026thinsp;12.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e76.9\u0026thinsp;\u0026plusmn;\u0026thinsp;12.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e76.6\u0026thinsp;\u0026plusmn;\u0026thinsp;12.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e78.4\u0026thinsp;\u0026plusmn;\u0026thinsp;12.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e76.8\u0026thinsp;\u0026plusmn;\u0026thinsp;12.2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDialysate temperature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e36.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e36.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e36.1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e36.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDialysate conductivity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e14.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e14.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e14.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlood flow rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e228.5\u0026thinsp;\u0026plusmn;\u0026thinsp;78.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e228.0\u0026thinsp;\u0026plusmn;\u0026thinsp;37.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e234.8\u0026thinsp;\u0026plusmn;\u0026thinsp;39.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e223.5\u0026thinsp;\u0026plusmn;\u0026thinsp;34.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e233.3\u0026thinsp;\u0026plusmn;\u0026thinsp;39.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUltrafiltration rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.70\u0026thinsp;\u0026plusmn;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.70\u0026thinsp;\u0026plusmn;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.70\u0026thinsp;\u0026plusmn;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.60\u0026thinsp;\u0026plusmn;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.70\u0026thinsp;\u0026plusmn;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVolume of ultrafiltration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.20\u0026thinsp;\u0026plusmn;\u0026thinsp;2.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.20\u0026thinsp;\u0026plusmn;\u0026thinsp;10.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.20\u0026thinsp;\u0026plusmn;\u0026thinsp;0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.10\u0026thinsp;\u0026plusmn;\u0026thinsp;0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.20\u0026thinsp;\u0026plusmn;\u0026thinsp;0.90\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLast SBP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e127.1\u0026thinsp;\u0026plusmn;\u0026thinsp;23.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e126.9\u0026thinsp;\u0026plusmn;\u0026thinsp;23.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e127.3\u0026thinsp;\u0026plusmn;\u0026thinsp;23.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e133.2\u0026thinsp;\u0026plusmn;\u0026thinsp;22.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e128.1\u0026thinsp;\u0026plusmn;\u0026thinsp;23.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLast DBP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e74.4\u0026thinsp;\u0026plusmn;\u0026thinsp;14.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e74.2\u0026thinsp;\u0026plusmn;\u0026thinsp;14.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e74.8\u0026thinsp;\u0026plusmn;\u0026thinsp;14.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e77.5\u0026thinsp;\u0026plusmn;\u0026thinsp;16.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e75.2\u0026thinsp;\u0026plusmn;\u0026thinsp;15.2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLast heart rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e76.7\u0026thinsp;\u0026plusmn;\u0026thinsp;13.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e76.8\u0026thinsp;\u0026plusmn;\u0026thinsp;13.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e76.0\u0026thinsp;\u0026plusmn;\u0026thinsp;13.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e75.6\u0026thinsp;\u0026plusmn;\u0026thinsp;13.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e75.9\u0026thinsp;\u0026plusmn;\u0026thinsp;13.3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eSBP, systolic blood pressure; DBP, diastolic blood pressure\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eEvaluation of Machine Learning Models and Selection of IDH Definition for Clinical Intervention Based on Predictive Performance and Clinical Feasibility\u003c/b\u003e \u003c/p\u003e \u003cp\u003eAll predictions were evaluated under the best F1 score and proper F1 score for practice. During the period from January 1, 2021 to December 31, 2021 (with a total of 302 dialysis days), the occurrence of IDH in Test-Set 1 under different IDH definitions were discrepant. Among them, the occurrence of IDH3 (51312) was much higher than those of IDH1 (10072) and IDH2(9338) (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). In Test-Set 1, the daily number of IDH predicted events was 30, 29, and 163 according to the definitions of IDH1, IDH2, and IDH3 using XGBoost, respectively (LightGBM: 28, 30, 163; RNN: 32, 26, 147, Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe occurrence of the three IDH definitions in Test Set 1 and the average daily IDH predicted values of each model at the best F1 score.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe occurrence of IDH\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThe average daily occurrence of IDH\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAverage number of daily IDH predictions\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eIDH1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e10072\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e33.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLightGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e28\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e32\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eIDH2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e9338\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e30.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e29\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLightGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e26\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eIDH3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e51312\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e169.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e163\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLightGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e163\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e147\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eIDH, Intradialytic hypotension; XGBoost, Extreme Gradient Boosting; LightGBM, Light Gradient Boosting Machine; RNN, recurrent neural network.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eWe used Test-Set 1, Test-Set 2 and Test-Set 3 to test the performance of the prediction model. Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e summarizes the basic parameters of different ML models under the three IDH definitions at the best threshold of the ROC-AUC. Under the definitions of IDH1 and IDH2, the ROC results of the three algorithm models were concentrated in the range of 0.906 to 0.920, with minor variation (less than 0.01). In Test-Set 2, the LightGBM model had the highest ROC-AUC for predicting IDH1 (0.915, 95%CI:0.911\u0026ndash;0.918). The Test-Set2/Test-Set1 PR-AUC ratios of IDH1 (RNN: 0.843, XGBoost: 0.892, LightGBM: 0.891) are higher than those of IDH2 (RNN: 0.808, XGBoost: 0.858, LightGBM: 0.859). Similarly, compared with Test-Set 1, the PR-AUC ratios of IDH1 in Test-Set 3 was greater than those of IDH2(Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The IDH1 was selected as the IDH definition for clinical intervention based on clinical feasibility and the results of internal validation and time-split validation.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eChanges in the ROC-AUC and PR-AUC of IDH1 and IDH2 in different IDH prediction models.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDATA SET\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eROC-AUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePR-AUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePR-AUC Ratio\u003c/p\u003e \u003cp\u003e(Test-Set2/Test-Set1 or Test-Set3 /Test-Set1)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIDH1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.914 (0.913 to 0.916)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.517 (0.515 to 0.520)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.906 (0.902 to 0.910)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.436 (0.429 to 0.443)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.914 (0.913 to 0.916)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.511 (0.508 to 0.513)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.988\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIDH2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.915 (0.913 to 0.916)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.506 (0.503 to 0.508)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.908 (0.904 to 0.912)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.409 (0.403 to 0.416)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.808\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.915 (0.914 to 0.917)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.498 (0.496 to 0.501)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.984\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIDH1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.914 (0.912 to 0.915)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.498 (0.496 to 0.501)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.914 (0.910 to 0.917)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.444 (0.438 to 0.450)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.892\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.915 (0.913 to 0.916)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.494 (0.492 to 0.496)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.992\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIDH2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.919 (0.918 to 0.921)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.499 (0.496 to 0.501)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.918 (0.914 to 0.921)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.428 (0.422 to 0.434)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.858\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.920 (0.919 to 0.921)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.493 (0.491 to 0.495)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.988\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLightGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIDH1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.913 (0.912 to 0.914)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.505 (0.503 to 0.508)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.915 (0.911 to 0.918)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.450 (0.444 to 0.456)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.891\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.914 (0.913 to 0.915)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.501 (0.498 to 0.503)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.992\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIDH2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.917 (0.915 to 0.918)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.498 (0.496 to 0.501)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.914 (0.910 to 0.917)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.428 (0.422 to 0.434)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.859\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest-Set3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.917 (0.916 to 0.919)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.493 (0.490 to 0.495)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.990\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eIDH1, intradialytic hypotension defined as SBP\u0026thinsp;\u0026lt;\u0026thinsp;90 mmHg when predialysis systolic blood pressure\u0026thinsp;\u0026lt;\u0026thinsp;160 mmHg (or SBP\u0026thinsp;\u0026lt;\u0026thinsp;100 mmHg when predialysis BP\u0026thinsp;\u0026ge;\u0026thinsp;160 mmHg); IDH2, intradialytic hypotension defined as nadir systolic BP\u0026thinsp;\u0026lt;\u0026thinsp;90 mmHg; RNN, recurrent neural network; XGBoost, Extreme Gradient Boosting; LightGBM, light gradient boosting machine; PR-AUC, precision‒recall area under the curve; ROC-AUC, receiver operating characteristic area under the curve.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn all IDH definitions, the ROC-AUC curves of the XGBoost, RNN, and LightGBM prediction models all rise steeply to the upper left corner indicates high sensitivity and low false positive rates (Figures \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e, S2, and S3). Model performance was also measured by calibration curves and reached the agreement between predicted and observed outcome (Figures S4). According to the evaluation of F1 score, the best performing predictive model is RNN, followed by XGBoost (Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). In view of the fact that the RNN model exhibits a high degree of sensitivity to timestamps, following systematic backend testing and a comprehensive assessment of its operational friendliness, tolerance for data missingness, and feasibility in clinical practice, the XGBoost model has been selected for subsequent clinical interventions. Based on the data from January 1, 2018, to July 24, 2023, the XGBoost model was retrained, and the optimal threshold was selected based on the highest F1-score (Table \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e). Ultimately, the XGBoost model with updated data chose a threshold of 0.3, where the ROC-AUC value was 0.933 and the PR-AUC value was 0.597 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eRCT Patient characteristics\u003c/h3\u003e\n\u003cp\u003eIn total, 32 patients were screened and randomized to the test group or placebo group. The first patient was enrolled on 2023/08/01, and the last patient was enrolled on 2023/08/20 (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents the baseline characteristics of patients stratified by randomization. The IDH event incidence rates in the AI-assisted group and the control group one year prior to enrollment were 39.5% (31.3%-77.6%) and 44.4% (36.8%-64.8%), respectively. No statistically significant differences were observed in baseline characteristics between the two groups.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline Characteristics in AI-assisted and Control Groups\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAI-assisted group\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eControl group\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender (male/female)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10/5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7/8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.269\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge(year)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e65.3\u0026thinsp;\u0026plusmn;\u0026thinsp;8.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e67.6\u0026thinsp;\u0026plusmn;\u0026thinsp;10.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.498\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDuration of dialysis (months)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e105.2\u0026thinsp;\u0026plusmn;\u0026thinsp;68.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e115.4\u0026thinsp;\u0026plusmn;\u0026thinsp;62.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.674\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertension(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9 (60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e9 (60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCardiovascular disease(%)*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6 (40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.427\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertension drug use(%)\u003csup\u003e#\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7 (46.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.245\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIncidence rate of IDH%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e39.5 (31.3, 77.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e44.4 (36.8, 64.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.455\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean SBP (mmHg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e111.1\u0026thinsp;\u0026plusmn;\u0026thinsp;16.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e111.3\u0026thinsp;\u0026plusmn;\u0026thinsp;20.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.979\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody mass index (kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22.3\u0026thinsp;\u0026plusmn;\u0026thinsp;3.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22.3\u0026thinsp;\u0026plusmn;\u0026thinsp;3.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.976\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003espKt/V\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.60\u0026thinsp;\u0026plusmn;\u0026thinsp;0.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.83\u0026thinsp;\u0026plusmn;\u0026thinsp;1.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.466\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNet UFV%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.84\u0026thinsp;\u0026plusmn;\u0026thinsp;0.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.97\u0026thinsp;\u0026plusmn;\u0026thinsp;0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.703\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIDWG%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4.21\u0026thinsp;\u0026plusmn;\u0026thinsp;1.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4.02\u0026thinsp;\u0026plusmn;\u0026thinsp;1.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.628\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSerum creatine (mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e987.1\u0026thinsp;\u0026plusmn;\u0026thinsp;164.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1013.4\u0026thinsp;\u0026plusmn;\u0026thinsp;185.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.684\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWBC (\u0026times;10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6.92\u0026thinsp;\u0026plusmn;\u0026thinsp;1.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5.92\u0026thinsp;\u0026plusmn;\u0026thinsp;1.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.070\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHemoglobin (g/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e115.5\u0026thinsp;\u0026plusmn;\u0026thinsp;10.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e118.9\u0026thinsp;\u0026plusmn;\u0026thinsp;9.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.380\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNT-proBNP (pg/mL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4636 (3468, 6036)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3000 (2318, 7648)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.534\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCRP (mg/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4.30 (2.40, 5.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5.20 (3.70, 9.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.067\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePTH (ng/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e389.5 (232.1, 469.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e220.4 (113.1, 385.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.051\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBUN (mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27.80\u0026thinsp;\u0026plusmn;\u0026thinsp;5.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28.64\u0026thinsp;\u0026plusmn;\u0026thinsp;4.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.659\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eβ2-Microglobulin (mg/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e33.12\u0026thinsp;\u0026plusmn;\u0026thinsp;3.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e34.10\u0026thinsp;\u0026plusmn;\u0026thinsp;4.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.541\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHematocrit%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e35.19\u0026thinsp;\u0026plusmn;\u0026thinsp;3.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e36.79\u0026thinsp;\u0026plusmn;\u0026thinsp;3.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.168\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlbumin (g/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e39.66\u0026thinsp;\u0026plusmn;\u0026thinsp;1.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e40.29\u0026thinsp;\u0026plusmn;\u0026thinsp;3.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.545\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePhosphate (mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.88\u0026thinsp;\u0026plusmn;\u0026thinsp;0.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.92\u0026thinsp;\u0026plusmn;\u0026thinsp;0.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.823\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003eIDH, intradialytic hypotension; SBP, systolic blood pressure; spKt/V, single pool Kt/V; UFV, ultrafiltration volume; IDWG, interdialytic weight gain; WBC, white blood cells; NT-proBNP, N-terminal pro-B-type natriuretic peptide; CRP, C-reactive protein; PTH, parathyroid hormone; BUN, blood urea nitrogen.\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003e*Cardiovascular disease was defined as the presence of either congestive heart failure or atherosclerotic cardiovascular disease, including: previous myocardial infarction, history of coronary artery angioplasty and/or stent placement,angina pectoris, evidence of coronary atherosclerotic disease, stroke, transient ischemic attacks, claudication, or aortic aneurysm.\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003e\u003csup\u003e#\u003c/sup\u003eMedications for treating hypertension should be discontinued on the day of dialysis and only used on non-dialysis days.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e\n\u003ch3\u003eThe Impact of AI-assisted on IDH Occurrence and IDH Incidence\u003c/h3\u003e\n\u003cp\u003eThe number of IDH occurrences in both groups during the observation period (pre-intervention), intervention period, and evaluation period (post-intervention) did not demonstrate any significant differences (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). From pre-intervention to intervention period, in the AI-assisted group, the reduction in the number of IDH occurrences was significantly greater than that observed in the control group (MD -8.13, 95% CI: -15.64 to -0.62, P\u0026thinsp;=\u0026thinsp;0.034) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA, Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). No significant impact of AI assistance on IDH incidence was found (Table S3).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe Impact of AI-assisted on IDH Occurrence\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026minus;\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAI-assisted group\u003c/p\u003e \u003cp\u003eMD (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl group\u003c/p\u003e \u003cp\u003eMD (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eBetween-Group difference MD (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePre-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e20.01 (11.47, 28.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e20.73 (12.19, 29.26)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c6\"\u003e \u003cp\u003e-0.72 (-13.15, 11.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.908\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntervention period\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e12.34 (3.81, 20.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e21.19 (12.66, 29.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c6\"\u003e \u003cp\u003e-8.86 (-21.29, 3.57)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.159\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePost-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11.14 (2.61, 19.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e15.66 (7.13, 24.19)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c6\"\u003e \u003cp\u003e-4.52 (-16.95, 7.91)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.469\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntervention period vs Pre-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-7.67 (-12.98, -2.36)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.47 (-4.84, 5.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.861\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c6\"\u003e \u003cp\u003e-8.13 (-15.64, -0.62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e0.034\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePost-intervention vs Pre-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-8.87 (-14.18, -3.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-5.07 (-10.38, 0.24)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.061\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c6\"\u003e \u003cp\u003e-3.80 (-11.31, 3.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.315\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePost-intervention vs Intervention period\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-1.20 (-6.51, 4.11)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.653\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-5.53 (-10.84, -0.22)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.041\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c6\"\u003e \u003cp\u003e4.33 (-3.18, 11.84)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.253\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eThe Impact of AI-assisted on SBP\u003c/h3\u003e\n\u003cp\u003eFollowing the intervention, the AI-assisted group experienced a significantly higher increase in SBP than the control group compared to the pre-intervention period (MD 5.96 mmHg, 95% CI: 0.02 to 11.90, P\u0026thinsp;=\u0026thinsp;0.049) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB, Table S4). During pre-intervention, intervention period and post-intervention period, there was no significant difference in the average SBP when IDH occurred in the two groups of patients (Table S5). Throughout the intervention period, including before and after its implementation, no statistically significant difference was observed in the average decrease of SBP between the two groups (Table S6). From pre-intervention to intervention period, the cumulative decrease in SBP within the AI-assisted group was significantly greater than that in the control group (MD -108.69, 95% CI: -209.83 to -7.56, P\u0026thinsp;=\u0026thinsp;0.036) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC, Table S7).\u003c/p\u003e\n\u003ch3\u003eThe Impact of AI-assisted on Patient Clinical Characteristics\u003c/h3\u003e\n\u003cp\u003eIn comparison to the pre-intervention period, the drop in pre-dialysis BUN in the AI-assisted group was significantly greater than that in the control group (MD -8.09, 95% CI: -11.32 to -4.85, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). Similarly, in AI-assisted group, compared to the pre-intervention period, the reduction in pre-dialysis serum creatine (Scr) was significantly greater than that in the control group (MD -116.84, 95% CI: -173.30 to -60.37, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). In the AI-assisted group, the reduction in white blood cell (WBC) count compared to the pre-intervention period was significantly greater than that observed in the control group (MD -1.15, 95% CI: -1.87 to -0.42, P\u0026thinsp;=\u0026thinsp;0.003). The variations of other clinical characteristics for the two groups before and after the intervention did not exhibit statistically significant differences (Table S8).\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe Impact of AI-assisted on Patient Clinical Characteristics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026minus;\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAI-assisted group\u003c/p\u003e \u003cp\u003eMD (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl group\u003c/p\u003e \u003cp\u003eMD (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBetween-Group difference MD (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePre-intervention\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e27.95 (25.04, 30.86)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e28.50 (25.61, 31.40)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.55 (-4.78, 3.68)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.790\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBUN\u003c/p\u003e \u003cp\u003e(mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePost-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22.21 (19.25, 25.17)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30.85 (27.90, 33.81)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e-8.64 (-12.95, -4.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.000\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePost-intervention vs Pre-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-5.74 (-8.02, -3.45)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.35 (0.06, 4.64)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e-8.09 (-11.32, -4.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePre-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e976.43 (889.15, 1063.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1023.07 (936.34, 1109.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e-46.64 (-173.70, 80.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.457\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSerum creatine (mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePost-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e833.86 (746.03, 921.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e997.34 (909.93, 1084.74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e-163.48 (-291.44, -35.51)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.014\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePost-intervention vs Pre-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-142.57 (-182.50, -102.64)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-25.73 (-65.66, 14.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e-116.84 (-173.30, -60.37)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePre-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7.03 (6.15, 7.91)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e5.81 (4.94, 6.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e1.22 (-0.07, 2.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.062\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWBC\u003c/p\u003e \u003cp\u003e(\u0026times;10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePost-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6.36 (5.48, 7.24)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6.29 (5.40, 7.18)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e0.07 (-1.22, 1.36)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.914\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePost-intervention vs Pre-intervention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-0.67 (-1.17, -0.17)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.48 (-0.04, 1.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e-1.15 (-1.87, -0.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.003\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eBUN, blood urea nitrogen; WBC, white blood cells\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we evaluate and identify the most suitable definition of IDH for AI-based prediction models. Analysis of three definitions\u0026mdash;IDH1 (dynamic threshold based on pre-dialysis SBP), IDH2 (fixed threshold), and IDH3 (dynamic change threshold)\u0026mdash;revealed distinct characteristics: IDH1 integrates logical symbol reasoning and timestamp attributes; IDH2 represents a single, extreme value that conforms to the underlying pathogenesis and follows binary logical symbol reasoning and decision tree characteristics; while IDH3, which reflects volume response to ultrafiltration during dialysis and reveals the contribution of timestamps to the defined variable, carries a risk of oversensitivity. Previous studies indicate that the frequent occurrence of IDH1 or IDH2 is significantly associated with all-cause mortality\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e, whereas IDH3 show no such correlation\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. However, the incidence of IDH3 (37.1%-39.2%) was significantly higher than that of IDH2 (2.7%-7.9%)\u003csup\u003e18, 19\u003c/sup\u003e. In this study, the predicted event rate of IDH3 (147\u0026ndash;163 events/day) substantially exceeded those of IDH1 (28\u0026ndash;32 events/day) and IDH2 (26\u0026ndash;30 events/day), confirming that its high sensitivity may lead to overtreatment risks.\u003c/p\u003e \u003cp\u003ePrevious research on AI-based IDH prediction has primarily utilized retrospective designs, relying on routine data from electronic medical record systems and dialysis equipment\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. While these studies involved multi-center cohorts with large sample sizes, they generally lacked a systematic comparison between internal validation and time-split validation. This limitation introduces uncertainty regarding the models' stability and generalizability in real-world clinical practice\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. In contrast, our study introduces a novel evaluation framework designed to assess temporal validity through a time-split validation approach. Using a prospectively collected, independent dataset, we systematically evaluated the models' performance in predicting outcomes for new patient populations across different time periods within the same institution. The results show that for both IDH1 and IDH2, the three algorithms maintained excellent ROC-AUC values, ranging from 0.90 to 0.92 across all test sets. The difference between internal validation (Test-Set 1) and time-split validation (Test-Set 2) was less than 0.01, demonstrating strong robustness.\u003c/p\u003e \u003cp\u003eSpecifically, when using the IDH1 definition, the ratio of PR-AUC between test set 2 and test set 1 was significantly higher than that under the IDH2 definition. Additionally, Test-Set 3 was designed to simulate a dynamic clinical environment with continuous patient enrollment and iterative model updates, closely mimicking real-world hemodialysis center operations. Compared to Test-Set 2, Test-Set 3 not only reflects the natural evolution of patient populations but also supports a more flexible and practical model-updating mechanism. Under this dynamic validation framework, the PR-AUC ratio advantage of the IDH1 definition over Test-Set1 was further amplified, providing practical evidence that the IDH1-based model exhibits enhanced robustness and adaptability in real-world applications.\u003c/p\u003e \u003cp\u003eA randomized controlled trial ultimately validated the clinical value of the IDH1 definition. The AI-assisted intervention group showed a significantly greater reduction in the number of IDH occurrences (MD -8.13, P\u0026thinsp;=\u0026thinsp;0.034) and a larger cumulative decrease in SBP (108.69 mmHg, P\u0026thinsp;=\u0026thinsp;0.036) compared to the control group. This clinical efficacy was supported by the exceptional discriminative ability of the XGBoost model under the IDH1 definition (ROC-AUC\u0026thinsp;=\u0026thinsp;0.933). In contrast to existing symptom-dependent definitions (e.g., K/DOQI 2005 \u003csup\u003e15\u003c/sup\u003e and KDIGO 2019 \u003csup\u003e21\u003c/sup\u003e), IDH1 enables standardized monitoring through objective parameters, not only improving diagnostic timeliness but also establishing a quantifiable benchmark for quality management in hemodialysis. This study demonstrates that the dynamically threshold-designed IDH1 definition significantly outperforms traditional definitions in both algorithmic robustness and clinical utility, providing a novel paradigm for precise IDH prevention and control.\u003c/p\u003e \u003cp\u003eWith regard to the predictive performance of IDH, our machine learning models achieved performance levels comparable to those reported in prior studies, while demonstrating distinct advantages. A retrospective study utilizing electronic medical records reported that the RNN model, when defining IDH by Nadir SBP\u0026thinsp;\u0026lt;\u0026thinsp;90 mmHg, achieved superior performance (ROC-AUC\u0026thinsp;=\u0026thinsp;0.94)\u003csup\u003e7\u003c/sup\u003e. Consistent with these findings, our RNN model exhibited the highest predictive accuracy (IDH2 Test3 ROC-AUC\u0026thinsp;=\u0026thinsp;0.915). Nevertheless, considering clinical applicability requirements for missing data handling, we pragmatically selected the secondary-performing XGBoost model for clinical implementation. Another study also defined Nadir90 as the IDH threshold and compared eight machine learning algorithms, ultimately determining XGBoost as the optimal predictor (ROC-AUC\u0026thinsp;=\u0026thinsp;0.936)\u003csup\u003e22\u003c/sup\u003e. Our research results confirm this conclusion, with the XGBoost model demonstrating robust performance (ROC-AUC\u0026thinsp;=\u0026thinsp;0.920 for IDH2 Test 3). Notably, a comparative analysis of IDH1 prediction revealed significant research gaps - the existing literature only records one multicenter retrospective study indicating that the performance of logistic regression, random forest, XGBoost, and deep learning models was comparable (ROC-AUC for XGBoost\u0026thinsp;=\u0026thinsp;0.87)\u003csup\u003e11\u003c/sup\u003e. Our XGBoost model shows significant improvement in this area, achieving higher predictive accuracy (ROC-AUC\u0026thinsp;=\u0026thinsp;0.915 for IDH1 Test 3).\u003c/p\u003e \u003cp\u003eWe found that from the pre-intervention to the intervention period, the AI intervention reduced the number of IDH occurrences. This observation may be attributed to the timely detection and management of IDH, which allowed for the rapid infusion of physiological saline or dialysate, improving vascular replenishment and reducing the temporary osmotic gradient between intracellular and extracellular compartments during dialysis\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. Additionally, this may potentially be associated with the elongation of the corresponding dialysis period. Previous studies have revealed that increasing the treatment duration could be beneficial for patients with hemodynamic or cardiovascular instability, significantly lowering the incidence of IDH\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eFurthermore, the definition of IDH cannot reflect the cumulative damage to various organs caused by repeated IDH occurrences during each dialysis session\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e, while the cumulative decrease in SBP may better reflect the impact of IDH recurrence on patients. The cumulative decrease in SBP refers to the cumulative amount of reduction in SBP from the IDH threshold to the occurrence of IDH. The cumulative decline of SBP is employed to assess the degree of impairment caused by recurrent IDH to the patient. Our study showed that the AI-assisted group had a significantly better SBP cumulative decrease than the control group from the pre-intervention to the intervention period. This suggests that AI-assisted intervention may reduce the cumulative damage to organs caused by IDH recurrence in patients.\u003c/p\u003e \u003cp\u003eIn this study, although the number of IDH occurrences in the AI-assisted group was significantly lower than that in the control group (p\u0026thinsp;=\u0026thinsp;0.034), the difference in the IDH incidence between the two groups did not reach statistical significance (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). This discrepancy may be attributed to insufficient statistical power due to the limited sample size. Future research should prioritize multi-center, large-scale randomized controlled trials with expanded sample sizes to enhance statistical power, thereby validating the clinical utility of the proposed model in reducing IDH incidence.\u003c/p\u003e \u003cp\u003eOur research design, data density and data quality offer considerable advantages. First, we prespecified the study protocol and prospectively collected the data for the study design. Currently, all ML model studies predicting IDH are based on retrospective data\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Additionally, as most ML models use data from multiple sources, including machine data, electronic health records, and other sources, extracting and merging such multimodal data can be challenging\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. The performance and generalizability of the model may be affected by the poor quality of some of the data or incomplete data, leading to information bias. In contrast, a prospective research design can overcome this limitation. This model training method is different from those used in previous ML studies\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Furthermore, our server collects hemodialysis data every minute and prospectively collects BP measurements every 30 minutes, thereby increasing the data density per unit time, resulting in only a 30-minute interval between the training sets of our models. Compared with traditional hourly data, collecting BP data every 30 minutes also increases the data volume by 80%, providing a more accurate time window for preventive intervention and greater potential clinical value\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. In terms of data quality, we have strict criteria for patient inclusion, requiring a minimum of one year of outpatient dialysis at the study hospital. In our study, the quality of patient data was controlled by quality control specialists, and patients with poor data quality, such as poor compliance and a high degree of missing data, were excluded. High-quality data can be used as training data to enhance the model's ability, improve its effectiveness, reliability, and stability, reduce the risk of overfitting, and enhance its robustness and generalizability. Therefore, AI-assisted intervention, based on our model, can still effectively reduce the occurrences of IDH in this complex, nonlinear, and heterogeneous pathophysiological process of IDH.\u003c/p\u003e \u003cp\u003eCurrently, the means of achieving timely diagnosis and intervention for IDH rely on dialysis equipment with online blood volume monitoring functions, data feedback mechanisms, and support from healthcare professionals\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. However, such hardware devices are typically associated with substantial costs. Our model can be implemented in dialysis facilities with lower hardware requirements, enabling them to use ML models to optimize computational algorithms, relying only on basic dialysis equipment to achieve real-time blood pressure monitoring and IDH occurrence prediction alerts.\u003c/p\u003e \u003cp\u003eThis study has several limitations. Primarily, there might be a certain amount of missing data in the dataset. Additionally, the sample size of this study is relatively limited and was conducted in a single center, which may restrict the applicability of the results in other healthcare settings with distinct patient populations and resources. However, owing to the methodological design and inherent characteristics of the machine learning dataset, the proposed model exhibits strong scalability, effectively mitigating these limitations. It can be compatible with data from other dialysis centers and generate models that simulate the real local environment, thus providing guidance for early diagnosis and intervention by local IDH.\u003c/p\u003e \u003cp\u003eIn conclusion, our model can be used to predict real-time risk of IDH defined by different nadir SBP based on pre-HD SBP stratification, and AI assisted intervention based on this model can significantly reduce the occurrence of IDH.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStudy Design\u003c/h2\u003e \u003cp\u003eThis study is a combination of prospective real-world research and a randomized controlled trial (chictr.org.cn: ChiCTR2000036973). All participants provided written informed consent before the study began. We collected demographic data and hemodialysis records for each participant. The study protocol has been approved by the ethics review board of Huashan Hospital (affiliated to Fudan University) (IRB number: KY2021-609). Our study consists of two parts: 1) a prospective real-world study, including data collection, models establishment and systematic internal and time-split validation of the models (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e,\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e); 2) a randomized controlled study in which the model is applied for AI intervention (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn the data collection, model establishment, and internal and time-split validation of the models section, we enrolled adult patients with ESRD who received regular hemodialysis at the Hemodialysis Center of Huashan Hospital Affiliated to Fudan University, Baoshan Branch (January 1, 2018 - December 31, 2021). Eligible ESRD patients were at least 18 years old and had completed one year of uninterrupted outpatient dialysis at this hospital. Exclusion criteria are detailed in Supplemental Methods.\u003c/p\u003e \u003cp\u003eIn the prospective cohort (Cohort 1) enrolled between January 1, 2018 and December 31, 2020, blood pressure was monitored using a standard upper-arm cuff for each patient during every hemodialysis session. Measurements were recorded at 30-minute intervals, ensuring the acquisition of at least nine blood pressure readings per dialysis session (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e)\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. Data from Cohort 1 included demographic, clinical and laboratory data obtained from electronic health records and dialysis machine data are detailed in Supplemental Methods. Each record was treated as a basic unit (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Seventy-five percent of the data were randomly assigned to the training set, while the remaining 25% was used for validation set. We collected dialysis data of cohort 1 (original patients) during the treatment process from January 1, 2021, to December 31, 2021, for internal validation (Test-Set 1), including real-time monitoring blood pressure and prediction IDH in the new follow-up data. At the same time, we collected a new data Test-Set (Test-Set 2) prospectively for time-split validation of patients starting from January 1, 2021, to December 31, 2021 (cohort 2, new enrolled patients). The data of Test-Set 1 and Test-Set 2 were combined to form Test-Set 3, which maybe in accordance with clinical practice.\u003c/p\u003e \u003cp\u003eIn AI intervention part, we recruited adult HD patients (2022/07-2023/07) receiving routine hemodialysis at the same Hospital and in the initial model training data set. Participants should be able to tolerate high-flux hemodialysis (HFHD) and Fresenius FX80 dialyzers (Fresenius Medical Care, Bad Homburg, Germany). The incidence of IDH in participants must be at least 30% within one year prior to enrollment. Other inclusion and exclusion criteria are described in Supplemental Methods.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eDefinition and Classification of IDH\u003c/h2\u003e \u003cp\u003e \u003cstrong\u003eDefinition 1\u003c/strong\u003e \u003cp\u003e(IDH1): intradialysis SBP\u0026thinsp;\u0026lt;\u0026thinsp;90 mmHg when the predialysis SBP\u0026thinsp;\u0026lt;\u0026thinsp;160 mmHg or intradialysis SBP\u0026thinsp;\u0026lt;\u0026thinsp;100 mmHg when the predialysis SBP\u0026thinsp;\u0026ge;\u0026thinsp;160 mmHg \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Definition 2 (IDH2): A nadir intradialysis SBP\u0026thinsp;\u0026lt;\u0026thinsp;90 mmHg \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Definition 3 (IDH3): A reduction in SBP of \u0026ge;\u0026thinsp;20 mmHg from predialysis to intradialysis levels \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. We regarded each IDH criterion as an independent binary outcome (0 or 1). The overall incidence of IDH was the proportion of hemodialysis sessions with IDH.\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eConstruction and Implementation of a Multi-Source Heterogeneous Hemodialysis Database for Intelligent Prediction of IDH\u003c/h2\u003e \u003cp\u003eThis study developed a machine learning-oriented hemodialysis database based on the commercially mature Hemodialysis Information System from GCSOFT Company Limited. The database integrates multi-source heterogeneous data, including patient diagnosis and treatment information related to hemodialysis, hemodialysis machine operational data, and laboratory test results interfaced with the hospital's Hospital Information System (HIS). Building upon this foundation, this study prospectively and innovatively established a structured data acquisition module.\u003c/p\u003e \u003cp\u003eObjective data primarily include pre-set or automatically collectable technical parameters and dynamic treatment process data, which are automatically uploaded to the server in real-time via system interfaces. Specifically, this includes: hemodialysis mode settings, ultrafiltration rates at various time points, periods of ultrafiltration pause, dialysate flow rate, conductivity, temperature, ultrafiltration mode, blood flow rate, isolated ultrafiltration periods, blood pressure and heart rate measured every 30 minutes using machine-coupled cuffs, machine-based fluid replenishment records, and records of non-machine-based fluid replenishment orders. All adjustments to medical orders or machine parameters are logged in real-time within the information system and synchronously uploaded to the server, ensuring data integrity and timeliness.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eMachine Learning Prediction Models and Evaluation Metrics\u003c/h2\u003e \u003cp\u003eWe explored three different ML prediction models: the recurrent neural network (RNN) model \u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e; Extreme Gradient Boosting (XGBoost) \u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e and Light Gradient Boosting Machine (LightGBM). This study employed a comprehensive set of metrics to evaluate model performance, including the F1 score, precision‒recall area under the curve (PR-AUC), and receiver operating characteristic area under the curve (ROC-AUC), among others. In our study, model calibration was performed using the Platt scaling method based on test set data. The Supplemental Methods provides detailed explanations of all evaluation metrics and the research framework for ML prediction models.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eRandomization and Intervention\u003c/h2\u003e \u003cp\u003eThrough a randomly generated sequence by the computer, patients were randomly assigned in a 1:1 ratio into two groups: one group received manual intervention after the ML model diagnosed IDH (AI-assisted group), and the other group received standard operation (the control group). The random sequence was generated by an independent database programmer prior to the trial to ensure blinding until the day before the trial. The first-line doctors were not involved in this study and had no knowledge of the random allocation.\u003c/p\u003e \u003cp\u003eAfter both groups of patients were enrolled, they would go through a 4-week observation/washout period (pre-intervention), a 4-week intervention period, and a 4-week evaluation period (post-intervention). During the entire 12-week trial phase, the AI system would monitor in real-time the blood pressure and dialysis parameters of all dialysis patients, and record the occurrence of IDH. In this study, ultra-pure dialysate was provided by the central dialysate delivery system (CDDS), and hemodialysis treatment was performed using GC-110N fully automated dialysis machines (JMS Co. Ltd., Tokyo, Japan). Hemodialysis mode was all HFHD.\u003c/p\u003e \u003cp\u003eDuring the observation/washout period, all patients who originally received hemodialysis/hemofiltration/hemoperfusion treatment were adjusted to receive three sessions of HFHD per week.\u003c/p\u003e \u003cp\u003eDuring the intervention period, when the AI system predicted an IDH event in the AI-assisted group at the next time point, a pop-up window for intervention prompting would appear on the computer in the nursing station (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The nurse instructed the patient to restrict their diet, offered dietary education, adjusted the temperature of the dialysis fluid to 35\u0026deg;C, and maintained the patient in Trendelenburg position as soon as possible. The nurse is required to administer 100 milliliters of ultrapure dialysate or normal saline intravenously. The duration of each dialysis session should be dynamically adjusted based on the patient's ultrafiltration rate. If necessary, the ultrafiltration rate can be moderately increased within the clinically acceptable range. In the control group, the AI system would predict IDH events, which were only recorded on the computer terminal without a window for intervention popping up. The intervention in the control group was based on clinical-driven operation, i.e., following the usual method when patients reported intolerance, described in Supplemental Methods.\u003c/p\u003e \u003cp\u003eDuring the evaluation period, the AI system would monitor blood pressure in real-time for both groups of patients and predict IDH events, only recorded on the computer terminal.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003ePrespecified Endpoints\u003c/h2\u003e \u003cp\u003ePrimary Endpoint: The number of IDH occurrences during hemodialysis sessions.\u003c/p\u003e \u003cp\u003eSecondary Endpoints: Included the incidence of IDH during hemodialysis, mean SBP during dialysis, mean SBP at IDH onset, mean reduction in SBP from the IDH threshold to the occurrence of IDH, and cumulative reduction in SBP from the IDH threshold to IDH occurrence.\u003c/p\u003e \u003cp\u003eExploratory Endpoint: Differences in clinical characteristics between the two groups post-intervention.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eData characteristics of randomized controlled trial\u003c/h2\u003e \u003cp\u003eDemographic, dialysis-related, and laboratory data were collected from enrolled patients during the study period, including gender, age, years on dialysis, the history of hypertension and cardiovascular disease, drugs used anti-hypertension, body mass index (BMI), Kt/V value, average weight gain during the dialysis interval relative to baseline, average net ultrafiltration relative to baseline, and creatinine level before dialysis, white blood cell (WBC) count, hemoglobin concentration, hematocrit, high-sensitivity C-reactive protein level, albumin concentration, electrolyte parameters (phosphorus), intact parathyroid hormone, and N-terminal pro-brain natriuretic peptide (NT-proBNP).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eThis study was designed as an Early Feasibility Study (EFS). The sample size of 32 patients was determined based on feasibility considerations to evaluate initial safety and device performance, in accordance with FDA guidelines for EFS, rather than a formal power calculation for confirmatory efficacy\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eBaseline characteristics of the experimental and control groups were summarized using descriptive statistics. Continuous variables were presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation or median with interquartile range for skewed distributions. Group comparisons at baseline were performed using the Student\u0026rsquo;s t-test or Wilcoxon rank-sum test, as appropriate. For the efficacy analysis, a linear mixed model for repeated measures was employed to compare the groups. The model included the baseline PTH level, time, treatment group, and the group-by-time interaction as fixed effects. A logarithmic transformation was applied to variables with skewed distributions prior to model fitting to satisfy normality assumptions. The treatment effect is expressed as the least-squares mean difference (LSMD) with its 95% confidence interval (CI). All analyses were performed using SAS 9.4..\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003eDisclosures\u003c/p\u003e\n\u003cp\u003eThere are no conflicts of interest to disclose.\u003c/p\u003e\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Scientific Research Project of Shanghai Municipal Health Commission, No. 201940271.\u003c/p\u003e\n\u003cp\u003eAuthor Contributions\u003c/p\u003e\n\u003cp\u003eYW, XZhu, JL, and CH conceived and designed the study. XZhou designed the clinical trial section. YW, XZhu, and JL provided data interpretation. JL, YH and XZhou conducted the data analysis. YW,XZhu and JL drafted and revised the article. YW, XZhu, JL, LY, XL, YR and JX were responsible for data collection. YW, LY, XL and YR performed patient recruitment. YW and JL configured and adjusted the logic and parameters of the machine learning model. YH and JL are responsible for machine learning coding and data maintenance. YH and JL tested the machine learning model. YT and ZJ are responsible for utilizing the program front-end and data maintenance. YW, XZhu and JL contributed equally to this work. All the authors reviewed and approved the manuscript for submission.\u003c/p\u003e\n\u003cp\u003eData Sharing Statement\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003eCode availability\u003c/p\u003e\n\u003cp\u003ehttps://github.com/WuYuanhaoSlug/IDH_Prediction\u003c/p\u003e\n"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eFlythe JE, Xue H, Lynch KE, Curhan GC, Brunelli SM (2015) Association of mortality risk with various definitions of intradialytic hypotension. J Am Soc Nephrol 26:724\u0026ndash;734\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAssimon MM, Flythe JE (2017) Definitions of intradialytic hypotension. Semin Dial 30:464\u0026ndash;472\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKuipers J et al (2019) The Prevalence of Intradialytic Hypotension in Patients on Conventional Hemodialysis: A Systematic Review with Meta-Analysis. Am J Nephrol 49:497\u0026ndash;506\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSands JJ et al (2014) Intradialytic hypotension: frequency, sources of variation and correlation with clinical outcome. Hemodial Int 18:415\u0026ndash;422\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang X et al (2022) Boosted machine learning model for predicting intradialytic hypotension using serum biomarkers of nutrition. Comput Biol Med 147:105752\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang X et al (2022) An optimized machine learning framework for predicting intradialytic hypotension using indexes of chronic kidney disease-mineral and bone disorders. Comput Biol Med 145:105510\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee H et al (2021) Deep Learning Model for Real-Time Prediction of Intradialytic Hypotension. Clin J Am Soc Nephrol 16:396\u0026ndash;406\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu YS et al (2021) Machine Learning Analysis of Time-Dependent Features for Predicting Adverse Events During Hemodialysis Therapy: Model Development and Validation Study. J Med Internet Res 23:e27098\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDong J et al (2023) Machine learning-based intradialytic hypotension prediction of patients undergoing hemodialysis: A multicenter retrospective study. Comput Methods Programs Biomed 240:107698\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang H et al (2023) Real-time prediction of intradialytic hypotension using machine learning and cloud computing infrastructure. Nephrol Dial Transpl 38:1761\u0026ndash;1769\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee H et al (2023) Prediction of intradialytic hypotension using pre-dialysis features-a deep learning-based artificial intelligence model. Nephrol Dial Transpl 38:2310\u0026ndash;2320\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang JC et al (2020) Predictive modeling of blood pressure during hemodialysis: a comparison of linear model, random forest, support vector regression, XGBoost, LASSO regression and ensemble method. Comput Methods Programs Biomed 195:105536\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEscobar GJ, Liu VX, Schuler A, Lawson B, Greene JD, Kipnis P (2020) Automated Identification of Adults at Risk for In-Hospital Clinical Deterioration. N Engl J Med 383:1951\u0026ndash;1960\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoftus TJ et al (2022) Artificial intelligence-enabled decision support in nephrology. Nat Rev Nephrol 18:452\u0026ndash;465\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eK/DOQI clinical (2005) practice guidelines for cardiovascular disease in dialysis patients. Am J Kidney Dis 45:S1\u0026ndash;153\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKooman J et al (2007) EBPG guideline on haemodynamic instability. Nephrol Dial Transpl 22(2):ii22\u0026ndash;44\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu Y, Lu J, Wang T, Zhu X, Xue J, You L (2024) Association of frequent intradialytic hypotension with the clinical outcomes of patients on hemodialysis: a prospective cohort study. Ren Fail 46:2296612\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRank N et al (2020) Deep-learning-based real-time prediction of acute kidney injury outperforms human predictive performance. NPJ Digit Med 3:139\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim HW et al (2022) Deep Learning Model for Predicting Intradialytic Hypotension Without Privacy Infringement: A Retrospective Two-Center Study. Front Med (Lausanne) 9:878858\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede Hond AAH, Shah VB, Kant IMJ, Van Calster B, Steyerberg EW, Hernandez-Boussard T (2023) Perspectives on validation of clinical predictive algorithms. NPJ Digit Med 6:86\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFlythe JE et al (2020) Blood pressure and volume management in dialysis: conclusions from a Kidney Disease: Improving Global Outcomes (KDIGO) Controversies Conference. \u003cem\u003eKidney Int\u003c/em\u003e 97, 861\u0026ndash;876\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang IN et al (2024) Personalized prediction of intradialytic hypotension in clinical practice: Development and evaluation of a novel AI dashboard incorporating risk factors from previous and current dialysis sessions. Int J Med Inf 190:105538\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArieff AI, Massry SG, Barrientos A, Kleeman CR (1973) Brain water and electrolyte metabolism in uremia: effects of slow and rapid hemodialysis. Kidney Int 4:177\u0026ndash;187\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKennedy AC, Linton AL, Eaton JC (1962) Urea levels in cerebrospinal fluid after haemodialysis. Lancet 1:410\u0026ndash;411\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilver SM, Sterns RH, Halperin ML (1996) Brain swelling after dialysis: old urea or new osmoles? Am J Kidney Dis 28:1\u0026ndash;13\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrunet P, Saingra Y, Leonetti F, Vacher-Coponat H, Ramananarivo P, Berland Y (1996) Tolerance of haemodialysis: a randomized cross-over trial of 5-h versus 4-h treatment time. Nephrol Dial Transpl 11(Suppl 8):46\u0026ndash;51\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLocatelli F, Manzoni C (1999) Duration of dialysis sessions\u0026ndash;was Hegel right? Nephrol Dial Transpl 14:560\u0026ndash;563\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReeves PB, Mc Causland FR, Mechanisms (2018) Clinical Implications, and Treatment of Intradialytic Hypotension. Clin J Am Soc Nephrol 13:1297\u0026ndash;1303\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLSQ NN (2023) A comparative study of the definitions of intradialytic hypotension correlated with increased mortality to identify universal predictors. Int J Med Inf 173:104975\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKotanko P, Zhang H, Wang Y (2023) Artificial Intelligence and Machine Learning in Dialysis: Ready for Prime Time? Clin J Am Soc Nephrol 18:803\u0026ndash;805\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin CJ, Chen YY, Pan CF, Wu V, Wu CJ (2019) Dataset supporting blood pressure prediction for the management of chronic hemodialysis. Sci Data 6:313\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSinha AD, Agarwal R (2009) Peridialytic, intradialytic, and interdialytic blood pressure measurement in hemodialysis patients. Am J Kidney Dis 54:788\u0026ndash;791\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen T, Guestrin C, XGBoost: (2016) A Scalable Tree Boosting System. In: \u003cem\u003eProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining\u003c/em\u003e). Association for Computing Machinery\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFDA. Investigational Device Exemptions (IDEs) for Early Feasibility Medical Device Clinical Studies, Including Certain First in Human (FIH) Studies.) (2013)\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"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":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Artificial intelligence, Hemodialysis, Intradialytic hypotension, Machine learning, AI-assisted intervention","lastPublishedDoi":"10.21203/rs.3.rs-8786359/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8786359/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIntradialytic hypotension (IDH) is associated with high mortality and morbidity. This study evaluates the application of machine learning (ML) for IDH diagnosis and management. This study consisted of a prospective real-world study and a pilot randomized controlled trial (RCT). Clinical data from 167 hemodialysis patients (2018–2020) were randomly divided into a training set (75%) and a validation set (25%). ML models (RNN, XGBoost, LightGBM) were assessed under three IDH definitions. The optimal XGBoost model, which utilized a stratified systolic blood pressure (SBP) threshold, achieved a receiver operating characteristic area under the curve (ROC-AUC) of 0.933, demonstrating robust predictive performance. In the RCT, 32 patients were allocated to AI-assisted IDH management or conventional care. Compared to controls, the AI-assisted group had a significantly greater reduction in IDH events (MD − 8.13, 95% CI: − 15.64 to − 0.62, P = 0.034) and a more marked improvement in cumulative SBP decline at IDH onset (MD − 108.69, 95% CI: − 209.83 to − 7.56, P = 0.036). The AI-assisted intervention, based on the XGBoost model predicting IDH risk using a stratified SBP threshold, significantly reduces IDH events, offering a novel strategy for the precise prevention and management of hypotension during dialysis.\u003c/p\u003e\n\u003cp\u003eClinical Trial registry name and registration number\u003c/p\u003e\n\u003cp\u003eResearch on Machine Learning-Based Information Systems for Predicting and Mitigating the Occurence of Intradialytic Hypotension, ChiCTR2000036973\u003c/p\u003e","manuscriptTitle":"Real-Time Prediction and Management of Intradialytic Hypotension with Machine Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-18 08:31:30","doi":"10.21203/rs.3.rs-8786359/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-communications","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"NCOMMS","sideBox":"Learn more about [Nature Communications](http://www.nature.com/ncomms/)","snPcode":"","submissionUrl":"https://mts-ncomms.nature.com/","title":"Nature Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Communications","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"9cfba788-8183-4eaf-ac2e-6159ba5342a9","owner":[],"postedDate":"March 18th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"This content is not available.","date":"2026-05-10T21:31:20+00:00","index":1,"fulltext":"This content is not available."}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":64424571,"name":"Health sciences/Nephrology/Renal replacement therapy/Haemodialysis"},{"id":64424572,"name":"Health sciences/Nephrology/Renal replacement therapy"}],"tags":[],"updatedAt":"2026-03-18T08:31:30+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-18 08:31:30","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8786359","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8786359","identity":"rs-8786359","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}