Machine learning models for predicting medium-term heart failure prognosis: Discrimination and calibration analyses

Machine learning models for predicting medium-term heart failure prognosis: Discrimination and calibration analyses | 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 Machine learning models for predicting medium-term heart failure prognosis: Discrimination and calibration analyses Takuya Nishino, Katsuhito Kato, Shuhei Tara, Daisuke Hayashi, and 12 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6008877/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The number of patients with heart failure (HF) is increasing with the aging population, shifting care from hospitals to clinics. Although predicting medium-term prognosis after discharge can enhance care and reduce readmissions, yet no established model has been evaluated for both discrimination and calibration. This multicenter study developed and validated machine learning (ML) models—including logistic regression, random forests, extreme gradient boosting, and light gradient boosting— to predict 180-day mortality or emergency hospitalization in 4,904 HF patients with HF. Patients were randomly split into training and validation sets (8:2), and models were trained and evaluated accordingly. All models showed acceptable performance based on the area under the precision-recall curve, good calibration according to the calibration slope and Brier score, and effective risk stratification. The SHapley Additive exPlanations algorithm identified nursing care needs as a key predictor alongside established laboratory values for HF prognosis. ML models effectively predict the 180-day prognosis patients with HF, with nursing care needs highlighting the importance of multidisciplinary collaboration. Clinical Trial Registration : URL: https://www.umin.ac.jp/ctr ; unique identifier: UMIN000054854 Biological sciences/Computational biology and bioinformatics Health sciences/Cardiology Health sciences/Health care Health sciences/Health care/Prognosis mortality rehospitalization receiver operating characteristic curve precision-recall curve calibration slope Brier score Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Heart failure (HF) is a global health challenge, affecting an estimated 64.3 million people worldwide [ 1 ] and leading to a growing number of hospitalizations [ 2 , 3 ] . This situation has raised concerns about strained healthcare provision and a marked increase in medical costs, leading to a shift in patient care for HF from hospitals to clinics [ 4 ] . In this context, effective cooperation between hospitals and clinics is essential. Routine patient care is delegated to primary care physicians in clinics, while hospital doctors conduct regular specialized checkups. Therefore, predicting the risk of medium-term deterioration after hospital discharge can assist primary care physicians in managing these patients. Furthermore, sharing this prediction among multiple healthcare professionals contributes to planning preventive care tailored to individual patients, thereby reducing the risk of readmission in patients with HF. Machine learning (ML) models developed by integrating multiple factors have been applied to predict outcomes in patients with HF and are expected to outperform conventional statistical methods in predictive accuracy [ 5 ] . Although many studies have demonstrated the utility of ML models in predicting short-term outcomes, such as morbidity, in-hospital mortality, and 30-day rehospitalization rates [ 6 – 8 ] , the effectiveness of these models in predicting medium- to long-term prognosis remains unclear. Previous ML models have predominantly relied on pathophysiological factors of the disease, such as clinical laboratory values, electrocardiographic features, echocardiographic findings, and medications [ 9 , 10 ] . However, these models have not accounted for information recorded by medical professionals other than doctors, including physical conditions, nursing care needs, medication adherence, and the social background of patients [ 2 , 11 ] . Comprehensive integration of these variables is crucial for improving medium-term prognostic accuracy. In this study, we aimed to develop and validate ML models that incorporate the physical status of patients in addition to clinical laboratory data and treatment details to predict the prognosis of patients with HF within 180 days of discharge. We selected ML models such as conventional logistic regression models with linear feature extraction and tree-based prediction models with nonlinear feature extraction. We evaluated the discrimination of these models by using the area under the receiver operating characteristic curve (AUROC), the area under the precision-recall curve (AUPRC), and the agreement between the models’ predicted probabilities and observed outcomes by calibration. Methods Study Design and Data Collection This multicenter cohort study was conducted using an inpatient database from four affiliated hospitals of Nippon Medical School, including Nippon Medical School Hospital, Musashi Kosugi Hospital, Tama Nagayama Hospital, and Chiba Hokusoh Hospital. The database was constructed using Diagnosis Procedure Combination (DPC) data and laboratory test values from medical records. Data for this study were accessed on July 21, 2024. The authors had no access to information that could identify individual participants during or after data collection. This study was approved by the Central Ethics Review Committee of Nippon Medical School (approval number M-2024-178) and conducted in accordance with the Declaration of Helsinki. Participant consent was obtained using an opt-out method. Reporting followed the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) guidelines [ 12 ] . Patient Selection and Endpoints Patients with HF in this study included those aged ≥ 18 years who were emergently hospitalized in the Department of Cardiovascular Medicine or the Cardiovascular Intensive Care Unit at one of the four Nippon Medical School-affiliated hospitals between April 2018 and September 2023. The eligibility criteria also required a brain natriuretic peptide (BNP) level of ≥ 100 pg/mL or an N-terminal pro-BNP (NT-proBNP) level of ≥ 300 pg/mL during hospitalization, in accordance with international definitions of HF [ 13 ] . Exclusions were made for patients with a hospital stay of < 5 days, those discharged owing to death or transfer to another hospital, or those lacking event occurrence with < 180 days of follow-up. The endpoint was a composite of all-cause mortality and emergency readmission within 180 days of discharge. Variables The DPC database collected data on all hospitalized patients, including demographics (age, sex), physical metrics height, weight, body mass index (BMI; calculated by dividing body weight [kg] by the square of height [m 2 ]), and clinical details (prior emergency hospitalization, comorbidities and procedures during hospitalization, and medications at discharge) [ 14 ] . The number of prior emergency hospitalizations was defined as those within 180 days after discharge. The medications at discharge recommended in the guidelines for HF and coronary artery disease were defined as those of guideline-directed medical therapy (GDMT) (Supplementary Table S1 ) [ 15 , 16 ] . Medications other than those included in GDMT were defined as those not included in the GDMT (ni-GDMT) medications. In this study, we focused on the number of ni-GDMT medications because they are considered to reflect the number of comorbidities, their severity, and polypharmacy. The patient’s physical status was evaluated using the nursing care needs score at discharge [ 14 ] . This score is the sum of the level of assistance required for turning over, transferring, oral hygiene, eating, and changing clothes (0 points: no assistance, 1 point: partial assistance, 2 points: full assistance), whether medical instructions were understood (0 points: yes, 1 point: no), and the presence of risky behavior (0 points: no, 2 points: yes) (Supplementary Table S2 ). Blood test data at admission and discharge were obtained from medical records and defined as the first sampling within 3 days of admission and the last sampling within 14 days before discharge, respectively. After applying the inclusion criteria, BNP was converted to NT-proBNP using the following conversion formula: NT-proBNP = 10 (1.1 × log10[BNP] + 0.57)[ 17 ] . Numerical data were scaled using the Standard Scaler. Imputation of Missing Data Missing data for BMI and blood test findings were imputed using a single imputation implemented in the Python MissForest package [ 18 , 19 ] . All variables were used for analysis because the missing rate for all variables was < 20%, which is the threshold for exclusion. Predictive Model Development We randomly divided the dataset into 80% training and 20% test sets using stratified sampling to preserve the endpoint occurrence rates of the original population. We used the following ML algorithms as predictive models: conventional logistic regression (LR), tree-based random forest (RF), extreme gradient boosting (XGB), and light gradient boosting machine (LGBM) [ 20 – 22 ] . In this study, we used recursive feature elimination with cross-validation (RFECV) of each model to select the optimal subset of features specific to each model. RFECV is a robust method that recursively eliminates less important features and builds a model with the remaining features to identify the optimal subset. Specifically, RFECV was implemented for each model using a 10-fold stratified cross-validation strategy to maximize the AUROC. After selecting the most relevant features, hyperparameter tuning was performed for each model using a grid search with cross-validation to determine the optimal model parameters. This step involved evaluating various combinations of hyperparameters and ultimately selecting the combination that maximized the AUROC for each model (Supplementary Table S3 ). As a result of the optimization process (Supplementary Table S4 ), the hyperparameter set with the highest AUROC score for each model was identified. The models were calibrated using isotonic regression [ 23 ] . SHapley Additive exPlanations (SHAP) values were used to explain the output of the ML models [ 24 ] . SHAP values quantify the contribution of each feature to the predictions made by the models, allowing for a better understanding of the factors driving the model’s decisions. Model Validation The bootstrap method was applied to the test set to evaluate the performance of the predictive models. A total of 2,500 bootstrap resamples were used to calculate 95% confidence intervals (CIs) for each performance metric. The discriminatory ability of the model was assessed using the AUROC and AUPRC [ 25 , 26 ] . For the calibration analysis, the predicted probabilities were divided into ten percentiles, and the mean predicted probability of the outcome and observed probability of the outcome for each bin were plotted. The following two indices were calculated to evaluate the calibration: the calibration slope, which indicates the agreement between the predicted probabilities and observations, and the Brier score, which measures the accuracy of the probability predictions. A calibration slope closer to 1 and a Brier score closer to 0 indicate ideal model performance. Risk Classification We classified the probabilities predicted by the ML models into three categories for risk stratification: Given that the probability of rehospitalization within 1 year for patients with HF is approximately 30% [ 27 , 28 ] , the categories were defined as low risk ( 0.15 and < 0.30), and high risk (≥ 0.30). Survival analysis was performed using the Kaplan–Meier method, and the log-rank test was used to compare survival distributions between the stratified groups. Package for Analysis All statistical analyses were performed using Python version 3.9.0 (Python Software Foundation, www.python.org ) and R software version 4.2.2 Patched (R Foundation for Statistical Computing, Vienna, Austria). A two-tailed test was performed, and a P-value of < 0.05 was considered statistically significant. Results Study Population and Baseline Characteristics Among the 9,519 patients, we excluded 905 who were discharged because of death, 1,402 who were transferred to other facilities, 466 with a hospital stay < 5 days, and 1,842 without events and a follow-up period < 180 days. Therefore, 4,904 participants were included in the final analysis, with 3,923 (80%) patients allocated to the training dataset and 981 (20%) allocated to the validation dataset (Fig. 1 ). Table 1 shows the descriptive statistics of the variables included in the training and test datasets before data processing, as well as the selected features. Outcomes occurred in 1,291 (26.3%) patients, and the features of the training and test datasets were well balanced. For the development of the ML models, out of a total of 61 variables, 28 features were selected for LR, 51 for RF, 52 for XGB, and 36 for LGBM (Supplementary Table S5 ). Table 1 Patient characteristics Variable Overall n = 4,904 Training dataset n = 3,923 Test dataset n = 981 Missing cases, n (%) Outcomes, n (%) 1291 (26.3) 1033 (26.3) 258 (26.3) 0 (0.0) Demographics Age (years) 77 (68–84) 77 (68–84) 77 (68–83) 0 (0.0) Male sex, n (%) 3142 (64.1) 2514 (64.1) 628 (64.0) 0 (0.0) Body mass index (kg/m 2 ) 22.8 (20.4–25.3) 22.8 (20.3–25.3) 22.8 (20.6–25.4) 128 (2.6) Hospitalization days (day) 16 (12–25) 16 (12–25) 17 (12–25) 0 (0.0) Nursing care needs score at discharge 1 (0–3) 1 (0–3) 1 (0–3) 0 (0.0) Prior emergency hospitalizations, n (%) 807 (16.5) 633 (16.1) 174 (17.7) 0 (0.0) Comorbidities Acute coronary syndrome: ICD10; I200, I21, n (%) 1,171 (23.9) 944 (24.1) 227 (23.1) 0 (0.0) Atrial fibrillation: ICD10; I48, n (%) 1,595 (32.5) 1,286 (32.8) 309 (31.5) 0 (0.0) Dyslipidemia: ICD10; E78, n (%) 2,621 (53.4) 2,102 (53.6) 519 (52.9) 0 (0.0) Diabetes mellitus: ICD10; E10–E14, n (%) 1,672 (34.1) 1,346 (34.3) 326 (33.2) 0 (0.0) Hypertension: ICD10; I10–I12, I15, n (%) 3,229 (65.8) 2,599 (66.3) 630 (64.2) 0 (0.0) Ischemic heart disease: ICD10; I201–209, I22–25, n (%) 1,261 (25.7) 1,007 (25.7) 254 (25.9) 0 (0.0) Medications at discharge Beta-blockers, n (%) 3,396 (69.2) 2,710 (69.1) 686 (69.9) 0 (0.0) ACE-Is/ARBs and ARNIs, n (%) 3,231 (65.9) 2,569 (65.5) 662 (67.5) 0 (0.0) Aspirin, n (%) 1,730 (35.3) 1,394 (35.5) 336 (34.3) 0 (0.0) P2Y12 inhibitors, n (%) 1,731 (35.3) 1,393 (35.5) 338 (34.5) 0 (0.0) Direct oral anticoagulants, n (%) 1,815 (37.0) 1,456 (37.1) 359 (36.6) 0 (0.0) Warfarin, n (%) 454 (9.3) 352 (9.0) 102 (10.4) 0 (0.0) Loop diuretics, n (%) 2,772 (56.5) 2,188 (55.8) 584 (59.5) 0 (0.0) Mineral corticoid receptor antagonists, n (%) 1,853 (37.8) 1,473 (37.5) 380 (38.7) 0 (0.0) Tolvaptan, n (%) 983 (20.0) 772 (19.7) 211 (21.5) 0 (0.0) SGLT2 inhibitors, n (%) 849 (17.3) 693 (17.7) 156 (15.9) 0 (0.0) Statins, n (%) 2,687 (54.8) 2,170 (55.3) 517 (52.7) 0 (0.0) Proton pump inhibitors, n (%) 1,766 (36.0) 1,414 (36.0) 352 (35.9) 0 (0.0) Pimobendan, n (%) 190 (3.9) 151 (3.8) 39 (4.0) 0 (0.0) Gout medications, n (%) 1,551 (31.6) 1,221 (31.1) 330 (33.6) 0 (0.0) Psychiatric drugs, n (%) 453 (9.2) 369 (9.4) 84 (8.6) 0 (0.0) Hypnotics, n (%) 1,465 (29.9) 1,164 (29.7) 301 (30.7) 0 (0.0) Number of ni-GDMT medications, n (%) 4 (2–7) 4 (2–7) 4 (2–7) 0 (0.0) In-hospital treatments Catecholamines, n (%) 680 (13.9) 550 (14.0) 130 (13.3) 0 (0.0) Opioids, n (%) 435 (8.9) 352 (9.0) 83 (8.5) 0 (0.0) Intravenous loop diuretics dosage (mg)/hospitalization 40 (0–200) 40 (0–200) 40 (0–220) 0 (0.0) Ventilator, n (%) 854 (17.4) 678 (17.3) 176 (17.9) 0 (0.0) Mechanical circulatory support, n (%) 246 (5.0) 196 (5.0) 50 (5.1) 0 (0.0) Hemodialysis, n (%) 227 (4.6) 185 (4.7) 42 (4.3) 0 (0.0) Laboratory findings at admission White blood cells (10 3 /µL) 7.6 (5.9–10.1) 7.7 (5.9–10.2) 7.4 (5.7–10.0) 20 (0.4) Hemoglobin (g/dL) 12.4 (10.7–14.1) 12.4 (10.6–14.1) 12.6 (10.8–14.3) 18 (0.4) Aspartate transaminase (IU/L) 29 (21–46) 29 (21–45) 31 (22–48) 176 (3.6) Alanine transaminase (IU/L) 21 (14–35) 21 (14–35) 22 (14–39) 148 (3.0) Albumin (g/dL) 3.7 (3.4–4.0) 3.7 (3.4–4.0) 3.7 (3.4–4.0) 76 (1.5) Blood urea nitrogen (mg/dL) 21.9 (16.2–31.8) 21.9 (16.3–31.7) 21.7 (16.0–32.5) 18 (0.4) Creatine kinase (IU/L) 105 (65–190) 106 (65–192) 101 (65–184) 49 (1.0) Uric acid (mg/dL) 6.2 (5.0–7.5) 6.20 (5–7.50) 6.1 (4.9–7.4) 405 (8.3) eGFR (mL/min/1.73 m 2 ) 46 (30–62) 46 (30–62) 47 (31–63) 18 (0.4) Sodium (mEq/L) 140 (137–142) 140 (137–142) 140 (137–142) 17 (0.3) Potassium (mEq/L) 4.2 (3.8–4.6) 4.2 (3.8–4.6) 4.2 (3.8–4.6) 18 (0.4) NT-proBNP (pg/mL) 2,713 (995–6708) 2,681 (985–6724) 2,863 (1021–6528) 167 (3.4) Laboratory findings at discharge White blood cells (10 3 /µL) 5.8 (4.7–7.0) 5.8 (4.7–7.1) 5.7 (4.7–6.9) 0 (0.0) Hemoglobin (g/dL) 11.9 (10.4–13.5) 11.8 (10.4–13.4) 12.0 (10.4–13.6) 3 (0.1) Aspartate transaminase (IU/L) 22 (17–28) 22 (17–28) 22 (18–29) 179 (3.6) Alanine transaminase (IU/L) 18 (12–27) 18 (12–27) 18 (12–28) 150 (3.1) Albumin (g/dL) 3.5 (3.2–3.8) 3.5 (3.2–3.8) 3.5 (3.2–3.8) 165 (3.4) Blood urea nitrogen (mg/dL) 20.2 (15.0–29.2) 20.1 (15.0–29.0) 20.2 (15.0–30.3) 2 (0.0) Creatine kinase (IU/L) 45 (31–68) 46 (31–68) 45 (31–68) 96 (2.0) Uric acid (mg/dL) 6.0 (4.8–7.3) 6.1 (4.9–7.3) 5.9 (4.7–7.2) 393 (8.0) eGFR (mL/min/1.73 m 2 ) 48 (33–63) 48 (32–63) 48 (33–63) 2 (0.0) Sodium (mEq/L) 140 (138–142) 140 (138–142) 140 (137–142) 0 (0.0) Potassium (mEq/L) 4.3 (4.0–4.6) 4.3 (4.0–4.6) 4.3 (4.0–4.6) 811 (16.5) NT-proBNP (pg/mL) 1,527 (717–3,530) 1,544 (723–3,544) 1,468 (672–3,506) 3 (0.1) Data are expressed as medians (interquartile ranges). ICD, International Statistical Classification of Diseases and Related Health Problems; ACE-Is, angiotensin-converting enzyme inhibitors; ARBs, angiotensin II receptor blockers; ARNIs, angiotensin receptor neprilysin inhibitors; SGLT2, sodium-glucose cotransporter 2; ni-GDMT, not included in guideline-directed medical therapy; eGFR, estimated glomerular filtration rate; NT-proBNP, N-terminal pro-brain natriuretic peptide Evaluation of Prediction Models After training the models, the AUROC was calculated based on the results of the predicted events in the test dataset using the trained ML models. The AUROC values for LR, RF, XGB, and LGBM were 0.734 (95% CI, 0.698–0.769), 0.753 (95% CI, 0.718–0.786), 0.744 (95% CI, 0.708–0.778), and 0.752 (95% CI, 0.716–0.785), respectively (Fig. 2 A). Consequently, the models were evaluated by AUPRC, which provides more informative insights into binary classification predictive models with imbalanced data, and the values for LR, RF, XGB, and LGBM were 0.505 (95% CI, 0.445–0.568), 0.513 (95% CI, 0.454–0.579), 0.524 (95% CI, 0.463–0.589), and 0.515 (95% CI, 0.454–0.582), respectively (Fig. 2 B). Calibration was assessed using a plot diagram of predicted probabilities and observations across the ten deciles of predicted risk, and favorable agreement between both values was observed in all models (Fig. 2 C). As calibration indices for the models, values for the calibration slope (LR, 0.964; RF, 1.046; XGB, 1.197; LGBM, 1.088) and Brier score (LR, 0.167; RF, 0.164; XGB, 0.165; LGBM, 0.164) were calculated (Table 2 ). Furthermore, the observed incidence of events was compared across the three defined risk categories (low, middle, and high risk). The observed event rates for each stratified group were within the predicted probability ranges (Table 3 ). Survival analysis using the Kaplan–Meier method demonstrated that the cumulative observed event rate significantly increased with risk stratification across the four models (Fig. 3 ). Table 2 Calibration indices Model Calibration slope (95% confidence interval) Brier score (95% confidence interval) Logistic regression 0.964 (0.804–1.120) 0.167 (0.154–0.181) Random forests 1.046 (0.890–1.260) 0.164 (0.151–1.177) Extreme gradient boosting 1.197 (1.013–1.377) 0.165 (0.153–0.178) Light gradient boosting machine 1.088 (0.927–1.250) 0.164 (0.152–0.177) Table 3 Observed outcomes for each risk stratum classified by predicted probability Risk stratum (range of predictive probability) Number of patients (total) Number of patients (outcomes) Observed outcome rate Logistic regression Low: <0.15 344 39 0.11 Middle: ≥0.15, < 0.30 241 48 0.20 High: ≥0.30 396 171 0.43 Random forests Low: <0.15 274 22 0.08 Middle: ≥0.15, < 0.30 350 70 0.20 High: ≥0.30 357 166 0.46 Extreme gradient boosting Low: <0.15 262 26 0.10 Middle: ≥0.15, < 0.30 347 65 0.19 High: ≥0.30 372 167 0.45 Light gradient boosting machine Low: <0.15 282 23 0.08 Middle: ≥0.15, < 0.30 317 66 0.21 High: ≥0.30 382 169 0.44 Model Interpretation The contribution of each feature to the prediction of outcomes was visualized using the SHAP algorithm. The features were ranked in descending order of importance scores, with red indicating high influence and blue indicating low influence. Established risk factors for HF, such as age, estimated glomerular filtration rate (eGFR), NT-proBNP levels, hemoglobin levels, albumin levels, and the number of previous hospitalizations, were ranked among the top contributors in all four models. Furthermore, nursing care needs scores and the number of ni-GDMT medications, which were the features we focused on in this study, were identified as highly influential factors for outcomes (Fig. 4 ). According to the SHAP analysis, the nursing care needs scores contributed 7.72% to the outcome prediction in LR, 4.50% in RF, 4.03% in XGB, and 5.92% in LGBM. The SHAP analysis of LGBM did not reveal clear, strong interactions between the nursing care needs scores and other features in the prediction, compared to the interactions observed between other features ( Fig. S1 ). Discussion In this multicenter retrospective cohort study, we developed ML models to predict all-cause mortality and emergency admission within 180 days of discharge in patients with HF. The models were trained on the training dataset and evaluated on the internal validation dataset. The discriminative power of each model fell within the range of AUROC (0.65–0.78) reported in previous studies on long-term prognosis prediction of patients with HF [ 6 ] . Moreover, the conventional LR model and tree-based models (RF, XGB, and LGBM) demonstrated favorable agreement between predicted and observed values in the calibration, and successful risk stratification was achieved in these models. Discrimination refers to the ability of a predictive model to distinguish correctly between positive and negative cases. In this study, the tree-based ML models demonstrated equal or better discriminatory power than the conventional LR model did for both AUROC and AUPRC. This is consistent with the concept that tree-based approaches, which make predictions based on nonlinear relationships, are more suitable than LR, which assumes linearity, especially because some features, such as eGFR and BMI, follow a U-shaped curve [ 29 , 30 ] . The ML models in this study demonstrated favorable agreement in calibration, which is an indicator of how closely the model’s predicted probability matched the observed outcome. Although this evaluation method with calibration is important for stratifying the risk of individual patients and applying predictive models to clinical practice [ 31 , 32 ] , there is a lack of model evaluation with calibration in a predictive model for HF [ 10 , 33 ] . In this study, the observed event rates in each stratum fell within the predicted probability ranges (low, middle, and high risk), suggesting that appropriate risk stratification for medium-term prognosis was achieved at the time of hospital discharge. In assessing the utility of a model, improperly structured calibration can be problematic because it leads to either over- or underestimation of risk [ 34 ] . Among the various algorithms reported as calibration methods for predictive models [ 35 ] , isotonic regression, which is one of the oldest and most commonly used methods, was applied in this study [ 23 ] . This approach ensured reasonable calibration was achieved, as indicated by the calibration plot, and effectively addressed the challenge of over- and underestimation of risk. However, emerging research being conducted on more precise calibration algorithms that utilize neural networks suggests the potential for improvement by applying these methods [ 36 ] . The nursing care needs score and the number of ni-GDMT medications highly influenced the prediction across all models, as shown by the SHAP analysis, in addition to well-known prognostic factors of HF [ 37 – 39 ] . This indicates that daily physical activity, comorbidity status, and polypharmacy are crucial in the prognosis of patients with HF. However, only the nursing care needs score was available to characterize physical activity in this study, as our retrospectively collected dataset did not include other variables related to the patient’s physical condition. Frailty and comorbidity burden are associated with rehospitalization within 6 months of discharge in older patients with HF [ 40 ] . Moreover, based on the recent knowledge of multiple comorbidities or social determinants of health, several physical activity parameters and social factors should be added to the features of ML models [ 41 , 42 ] . We constructed the dataset for this study using DPC and clinical laboratory data, which are automatically stored in an electronic format and can be readily exported. The use of DPC and electronic medical records has been standardized across Japanese university hospitals, and several clinical studies have utilized similar databases [ 43 ] . Therefore, our results are generalizable and can be easily confirmed using data from other hospitals for external validation. However, our dataset only included inpatients at a university hospital that provides advanced and specialized treatments and did not reflect comprehensive HF care in the community region with hospital–clinic cooperation. Incorporating clinical data from primary care physicians in clinics and patient information (such as social background, physical activity, and living conditions) recorded by healthcare professionals other than doctors into the database would lead to the construction of a more accurate model for predicting the progression of HF. Building such a comprehensive database with a regional hospital–clinic–care team will be the next challenge. Limitations This study has some limitations. First, as a retrospective study, the explanatory variables for the ML models were limited to readily available clinical data with few missing values. Data such as echocardiographic and electrocardiographic findings, vital signs, and social factors were not included in this study. Second, this study was conducted solely in healthcare facilities in Japan; therefore, the findings may not apply to patients in other regions. Future studies should validate the ML models using independent external data from various regions and hospitals. Third, only four ML algorithms were used. Other ML algorithms, such as neural networks and support vector machines, may exhibit better predictive performances. Finally, NT-proBNP levels ≥ 300 pg/mL and BNP levels ≥ 100 pg/mL were used to diagnose HF. However, data on the conversion between BNP and NT-proBNP suggest that the BNP level equivalent to an NT-proBNP level of 300 pg/mL is lower, at approximately 60–75 pg/mL [ 44 ] . This discrepancy might have led to the inaccurate inclusion of patients with HF. Additionally, because a formula for conversion between BNP and NT-proBNP was used for model development, patients assessed with BNP may not have been accurately evaluated. Conclusions For patients with HF, we developed ML models to predict all-cause mortality and emergency admission in the medium term within 180 days after hospital discharge. This study demonstrates that ML models have favorable discrimination and agreement between predicted probabilities and observations. These ML models can be useful in reducing rehospitalization after discharge through risk stratification in individual patients. Furthermore, the influence of nursing care needs on prediction indicates the importance of multidisciplinary collaboration in HF care. Declarations Competing interests The authors declare no competing interests. Author Contribution T N: Data Curation, Formal analysis, Software, Writing- Original draft preparation. K K: Conceptualization, Methodology, Data Curation, Writing- Original draft preparation. S T: Conceptualization, Data curation, Writing - Review & Editing, Project administration. D H: Conceptualization, Methodology, Writing - Review & Editing. T S: Supervision, Validation, Visualization, Writing - Review & Editing. T T: Validation, Visualization, Writing - Review & Editing. Y K: Conceptualization, Methodology, Resources. T Y: Data curation. M M: Data curation. E K: Data curation. N K: Data curation. A S: Visualization, Data curation. T O: Conceptualization, Supervision. S Y: Supervision. Y K: Supervision. K A: Supervision, Funding acquisition, Project administration. All authors reviewed the manuscript. Acknowledgement The authors extend their sincere thanks to all the people involved in patient care, including emergency staff, technicians, medical engineers, nurses, pharmacists, physicians, and surgeons at Nippon Medical School Hospital, Musashi-Kosugi Hospital, Tama Nagayama Hospital, Chiba Hokusoh Hospital, and Nippon Medical School. Data Availability The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. References GBD. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990–2017: A systematic analysis for the Global Burden of Disease Study 2017. Lancet 392 , 1789–1858 (2018). Yasuda, S., Miyamoto, Y. & Ogawa, H. Current status of cardiovascular medicine in the aging society of Japan. Circulation 138 , 965–967 (2018). Okura, Y. et al. Impending epidemic: Future projection of heart failure in Japan to the year 2055. Circ. J. 72 , 489–491 (2008). Di Salvo, T. G. & Stevenson, L. W. Interdisciplinary team-based management of heart failure. Dis. Manag Health Outcomes . 11 , 87–94 (2003). Błaziak, M. et al. An artificial intelligence approach to guiding the management of heart failure patients using predictive models: A systematic review. Biomedicines 10 , 2188 (2022). Croon, P. M. et al. Current state of artificial intelligence-based algorithms for hospital admission prediction in patients with heart failure: A scoping review. Eur. Heart J. Digit. Health . 3 , 415–425 (2022). Medhi, D. et al. Artificial intelligence and its role in diagnosing heart failure: A narrative review. Cureus 16 , e59661 (2024). Yu, M. Y. & Son, Y. J. Machine learning-based 30-day readmission prediction models for patients with heart failure: A systematic review. Eur. J. Cardiovasc. Nurs. 23 , 711–719 (2024). Miyagawa, S. et al. Japan heart failure model – Derivation and accuracy of survival prediction in Japanese heart failure patients. Circ. Rep. 1 , 29–34 (2018). Shin, S. et al. Machine learning vs. conventional statistical models for predicting heart failure readmission and mortality. ESC Heart Fail. 8 , 106–115 (2021). Georgiev, K. D., Hvarchanova, N., Georgieva, M. & Kanazirev, B. The role of the clinical pharmacist in the prevention of potential drug interactions in geriatric heart failure patients. Int. J. Clin. Pharm. 41 , 1555–1561 (2019). Collins, G. S. et al. TRIPOD + AI statement: Updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ 385 , q902 (2024). Bozkurt, B., Coats, A. & Tsutsui, H. Universal definition and classification of heart failure. J . Card . Fail . 27, S1071-9164(21)00050 – 6 (2021). Hayashida, K., Murakami, G., Matsuda, S. & Fushimi, K. History and profile of Diagnosis Procedure Combination (DPC): Development of a real data collection system for acute inpatient care in Japan. J. Epidemiol. 31 , 1–11 (2021). Tsutsui, H. et al. JCS/JHFS 2021 guideline focused update on diagnosis and treatment of acute and chronic heart failure. J. Card Fail. 27 , 1404–1444 (2021). Nakamura, M. et al. JCS 2020 guideline focused update on antithrombotic therapy in patients with coronary artery disease. Circ. J. 84 , 831–865 (2020). Fitzmaurice, G. M. & Laird, N. M. Multivariate analysis: Discrete variables (Logistic regression) in International Encyclopedia of the Social & Behavioral Science 10221–10228Elsevier, Amsterdam, (2001). Dong, W. et al. Generative adversarial networks for imputing missing data for big data clinical research. BMC Med. Res. Methodol. 21 , 78 (2021). Waljee, A. K. et al. Comparison of imputation methods for missing laboratory data in medicine. BMJ Open. 3 , e002847 (2001). Breiman, L. Mach. Learn. 45 , 5–32 (2001). Chen, T., Guestrin, C. & XGBoost A scalable tree boosting system in: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 785–794ACM, New York, USA, (2016). Machado, M. R., Karray, S., de Sousa, I. T. & LightGBM An effective decision tree gradient boosting method to predict customer loyalty in the finance industry. In: 14th International Conference on Computer Science & Education (ICCSE); vol 2019. 1111–1116IEEE Publs, Toronto, (2019). Zadrozny, B. & Elkan, C. Transforming classifier scores into accurate multiclass probability estimates in Proceedings of the 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining ; New York 694–699 (ACM Press; 2002). (2002). Lundberg, S. M. & Lee, S. I. A unified approach to interpreting model predictions in Adv . Neural Inf . Process . Syst . 30, 4768–4777 (2017). Harrell, F. E. Regression Modeling Strategies : With Applications to Linear Models , Logistic Regression , and Survival Analysis Springer, New York,. (2001). Saito, T. & Rehmsmeier, M. The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. PLOS One . 10 , e0118432 (2015). Ide, T. et al. Clinical characteristics and outcomes of hospitalized patients with heart failure from the large-scale Japanese Registry of Acute Decompensated Heart Failure (JROADHF). Circ. J. 85 , 1438–1450 (2021). Shiraishi, Y. et al. 9-year trend in the management of acute heart failure in Japan: A report from the national consortium of acute heart failure registries. J. Am. Heart Assoc. 7 , e008687 (2018). Zhang, J. et al. Body mass index and all-cause mortality in heart failure patients with normal and reduced ventricular ejection fraction: A dose-response meta-analysis. Clin. Res. Cardiol. 108 , 119–132 (2019). Watanabe, Y. et al. Fractional excretion of urea nitrogen can identify true worsening renal function in patients with heart failure. ESC Heart Fail. 11 , 2043–2054 (2024). Walsh, C. G., Sharman, K. & Hripcsak, G. Beyond discrimination: A comparison of calibration methods and clinical usefulness of predictive models of readmission risk. J. Biomed. Inf. 76 , 9–18 (2017). Abdul-Samad, K. et al. Comparison of machine learning and conventional statistical modeling for predicting readmission following acute heart failure hospitalization. Am. Heart J. 277 , 93–103 (2024). Christodoulou, E. et al. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J. Clin. Epidemiol. 110 , 12–22 (2019). Alba, A. C. et al. Discrimination and calibration of clinical prediction models: Users’ guides to the medical literature. JAMA 318 , 1377–1384 (2017). Huang, Y., Li, W., Macheret, F. & Gabriel, R. A. Ohno-Machado, L. A tutorial on calibration measurements and calibration models for clinical prediction models. J. Am. Med. Inf. Assoc. 27 , 621–633 (2020). Guo, C., Pleiss, G., Sun, Y. & Weinberger, K. Q. On calibration of modern neural networks in P roceedings of the 34th International Conference on Machine Learning . Sydney, Australia 1321–1330. (2017). Schaub, J. A. et al. Amino-terminal pro-B-type natriuretic peptide for diagnosis and prognosis in patients with renal dysfunction: A systematic review and meta-analysis. JACC Heart Fail. 3 , 977–989 (2015). Xia, H., Shen, H., Cha, W. & Lu, Q. The prognostic significance of anemia in patients with heart failure: A meta-analysis of studies from the last decade. Front. Cardiovasc. Med. 8 , 632318 (2021). Nishikido, T., Oyama, J. I., Nagatomo, D. & Node, K. A reduction of BMI predicts the risk of rehospitalization and cardiac death in non-obese patients with heart failure. Int. J. Cardiol. 276 , 166–170 (2019). Okoye, C. et al. Predicting mortality and re-hospitalization for heart failure: A machine-learning and cluster analysis on frailty and comorbidity. Aging Clin. Exp. Res. 35 , 2919–2928 (2023). Sterling, M. R. et al. Social determinants of health and 90-day mortality after hospitalization for heart failure in the REGARDS study. J. Am. Heart Assoc. 9 , e014836 (2020). Dickens, C., Dickson, V. V. & Piano, M. R. Perceived stress among patients with heart failure who have low socioeconomic status: A mixed-methods study. J. Cardiovasc. Nurs. 34 , E1–E8 (2019). Kanazawa, N. et al. Existing data sources for clinical epidemiology: Database of the National Hospital Organization in Japan. Clin. Epidemiol. 14 , 689–698 (2022). Kasahara, S. et al. Conversion formula from B-type natriuretic peptide to N-terminal proBNP values in patients with cardiovascular diseases. Int. J. Cardiol. 280 , 184–189 (2019). Additional Declarations No competing interests reported. Supplementary Files STable20250130.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6008877","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":414604457,"identity":"8f494756-d080-4370-b687-5300e20c9eb9","order_by":0,"name":"Takuya Nishino","email":"","orcid":"","institution":"Nippon Medical School","correspondingAuthor":false,"prefix":"","firstName":"Takuya","middleName":"","lastName":"Nishino","suffix":""},{"id":414604458,"identity":"7c8f9451-51fc-4bee-8767-ceec831ed663","order_by":1,"name":"Katsuhito Kato","email":"","orcid":"","institution":"Nippon Medical School","correspondingAuthor":false,"prefix":"","firstName":"Katsuhito","middleName":"","lastName":"Kato","suffix":""},{"id":414604459,"identity":"b761eadc-c485-4c2a-96c2-51a2af4d22d1","order_by":2,"name":"Shuhei Tara","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvklEQVRIiWNgGAWjYDAC+YONDz5UwHg8xGiRYD5sOOMMaVrY0oR520hy1+weMwbeeXei5RuYH35gkLlDWIvBnTNmDyS3PcvdcIDNWIKB5xkRWhhyzA0Mtx3O3cDAYAb0y2EiHNaQYyaROOdw7vwG9m/EaWG4kZYmcbDhcG7DAR4ibTE4c/iwYcMxoMMO8xRLJBDjF/n2xsbHf2qADmtv3/jhYw8RIYYAzECc2HOAFC1g8IN0LaNgFIyCUTD8AQAEdj5wp1HMuQAAAABJRU5ErkJggg==","orcid":"","institution":"Nippon Medical School","correspondingAuthor":true,"prefix":"","firstName":"Shuhei","middleName":"","lastName":"Tara","suffix":""},{"id":414604460,"identity":"34d295ef-1967-4a8f-b6eb-381c3865db7d","order_by":3,"name":"Daisuke Hayashi","email":"","orcid":"","institution":"Nippon Medical School Hospital","correspondingAuthor":false,"prefix":"","firstName":"Daisuke","middleName":"","lastName":"Hayashi","suffix":""},{"id":414604461,"identity":"bd183171-a76b-45f8-adb0-1efcf2da6d3d","order_by":4,"name":"Tomohisa Seki","email":"","orcid":"","institution":"The University of Tokyo Hospital","correspondingAuthor":false,"prefix":"","firstName":"Tomohisa","middleName":"","lastName":"Seki","suffix":""},{"id":414604462,"identity":"81adc94c-d690-4ae2-9122-9c1b7664456f","order_by":5,"name":"Toru Takiguchi","email":"","orcid":"","institution":"The University of Tokyo Hospital","correspondingAuthor":false,"prefix":"","firstName":"Toru","middleName":"","lastName":"Takiguchi","suffix":""},{"id":414604463,"identity":"992adf3d-8556-4f7f-a72d-58f03b94f3c5","order_by":6,"name":"Yoshiaki Kubota","email":"","orcid":"","institution":"Nippon Medical School","correspondingAuthor":false,"prefix":"","firstName":"Yoshiaki","middleName":"","lastName":"Kubota","suffix":""},{"id":414604464,"identity":"86453e99-d406-45f2-9fcb-19d7e4de8202","order_by":7,"name":"Takeshi Yamamoto","email":"","orcid":"","institution":"Nippon Medical School Hospital","correspondingAuthor":false,"prefix":"","firstName":"Takeshi","middleName":"","lastName":"Yamamoto","suffix":""},{"id":414604465,"identity":"099f1359-46ec-457d-8214-01a9ab10b828","order_by":8,"name":"Mitsunori Maruyama","email":"","orcid":"","institution":"Nippon Medical School Musashi-Kosugi Hospital","correspondingAuthor":false,"prefix":"","firstName":"Mitsunori","middleName":"","lastName":"Maruyama","suffix":""},{"id":414604466,"identity":"81e43223-9021-4848-9f39-2663a1aad683","order_by":9,"name":"Eitaro Kodani","email":"","orcid":"","institution":"Nippon Medical School Tama Nagayama Hospital","correspondingAuthor":false,"prefix":"","firstName":"Eitaro","middleName":"","lastName":"Kodani","suffix":""},{"id":414604467,"identity":"6004bc87-6821-496c-9ae0-ef7802ff3abd","order_by":10,"name":"Nobuaki Kobayashi","email":"","orcid":"","institution":"Nippon Medical School Chiba Hokusoh Hospital","correspondingAuthor":false,"prefix":"","firstName":"Nobuaki","middleName":"","lastName":"Kobayashi","suffix":""},{"id":414604468,"identity":"842e7f7b-0c51-4a9d-968e-120362f9497a","order_by":11,"name":"Akihiro Shirakabe","email":"","orcid":"","institution":"Nippon Medical School Chiba Hokusoh Hospital","correspondingAuthor":false,"prefix":"","firstName":"Akihiro","middleName":"","lastName":"Shirakabe","suffix":""},{"id":414604469,"identity":"d4b800a8-f731-4a2e-8295-e9f45078b96e","order_by":12,"name":"Toshiaki Otsuka","email":"","orcid":"","institution":"Nippon Medical School","correspondingAuthor":false,"prefix":"","firstName":"Toshiaki","middleName":"","lastName":"Otsuka","suffix":""},{"id":414604470,"identity":"63b21dcd-0856-4266-a461-be9b16df5b49","order_by":13,"name":"Shoji Yokobori","email":"","orcid":"","institution":"Nippon Medical School","correspondingAuthor":false,"prefix":"","firstName":"Shoji","middleName":"","lastName":"Yokobori","suffix":""},{"id":414604471,"identity":"146bf733-204f-4df8-9f7f-dc816495ab85","order_by":14,"name":"Yukihiro Kondo","email":"","orcid":"","institution":"Nippon Medical School","correspondingAuthor":false,"prefix":"","firstName":"Yukihiro","middleName":"","lastName":"Kondo","suffix":""},{"id":414604472,"identity":"c3c5a763-d8ba-4319-ba35-f475a0cb7414","order_by":15,"name":"Kuniya Asai","email":"","orcid":"","institution":"Nippon Medical School","correspondingAuthor":false,"prefix":"","firstName":"Kuniya","middleName":"","lastName":"Asai","suffix":""}],"badges":[],"createdAt":"2025-02-11 15:53:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6008877/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6008877/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":76996549,"identity":"de49fc02-c42d-441e-b4c9-55ee292e805a","added_by":"auto","created_at":"2025-02-24 06:48:29","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":106848,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart for patient assignment and model development\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBNP, brain natriuretic peptide; NT-proBNP, N-terminal pro-brain natriuretic peptide; RFECV, recursive feature elimination with cross-validation\u003c/p\u003e","description":"","filename":"Slide1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6008877/v1/09a18bd71d1fefed04cc9817.jpg"},{"id":76997415,"identity":"e8327988-e1d1-428f-bfc4-de4996e534a7","added_by":"auto","created_at":"2025-02-24 06:56:28","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":71893,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eModel validations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(\u003cstrong\u003ea\u003c/strong\u003e) The receiver operating characteristics curve and area under the receiver operating characteristic (AUROC) curve of the models. AUROC is presented with a 95% confidence interval (CI). (\u003cstrong\u003eb\u003c/strong\u003e) The precision-recall curves and area under the precision-recall (AUPRC) curve of the models. The AUPRC is presented with a 95% CI. (\u003cstrong\u003ec\u003c/strong\u003e) Calibration plot of the models. LR, logistic regression; RF, random forests; XGB, extreme gradient boosting; LGBM, light gradient boosting machine\u003c/p\u003e","description":"","filename":"Slide2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6008877/v1/c1721e211631a111dce172ff.jpg"},{"id":76996550,"identity":"9a8455a5-eb06-4246-b101-2ba37a1a8504","added_by":"auto","created_at":"2025-02-24 06:48:29","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":81016,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eKaplan–Meier curve analysis of 180-day all-cause mortality and emergency readmission\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe cumulative incidence of events observed in each model is compared across the three risk categories (low, middle, and high risk).\u003c/p\u003e","description":"","filename":"Figure320250204.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6008877/v1/73b267bd87e4315be4c358a2.jpg"},{"id":76997416,"identity":"6f4e2089-3fc0-47a1-adbc-2ab7192ca623","added_by":"auto","created_at":"2025-02-24 06:56:29","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":132327,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHapley Additive exPlanation (SHAP) values in the models\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe top 20 variables of SHAP analysis in each model are shown, along with the impact of their contribution to the prediction. eGFR, estimated glomerular filtration rate; ni-GDMT, not included in the guideline-directed medical therapy; ACE-I, angiotensin-converting enzyme inhibitor; ARB, angiotensin II receptor blocker; ARNI, angiotensin receptor neprilysin inhibitor; ALT, alanine aminotransferase; NT-proBNP, N-terminal pro-brain natriuretic peptide; AST, aspartate aminotransferase\u003c/p\u003e","description":"","filename":"Figure420250204.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6008877/v1/1973958e9482cdc26ac2968b.jpg"},{"id":93036540,"identity":"3d1d4ef8-8e0d-4b03-a92b-8ffd2cd2b025","added_by":"auto","created_at":"2025-10-08 11:36:28","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1809334,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6008877/v1/13ceef89-d2c1-4163-9e66-d27d770a0d05.pdf"},{"id":76996559,"identity":"bd466972-5d11-436b-9585-40a394ec7d6f","added_by":"auto","created_at":"2025-02-24 06:48:29","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":259451,"visible":true,"origin":"","legend":"","description":"","filename":"STable20250130.docx","url":"https://assets-eu.researchsquare.com/files/rs-6008877/v1/e5ca5169464df7327a70af0f.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine learning models for predicting medium-term heart failure prognosis: Discrimination and calibration analyses","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHeart failure (HF) is a global health challenge, affecting an estimated 64.3\u0026nbsp;million people worldwide\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e and leading to a growing number of hospitalizations\u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. This situation has raised concerns about strained healthcare provision and a marked increase in medical costs, leading to a shift in patient care for HF from hospitals to clinics\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. In this context, effective cooperation between hospitals and clinics is essential. Routine patient care is delegated to primary care physicians in clinics, while hospital doctors conduct regular specialized checkups. Therefore, predicting the risk of medium-term deterioration after hospital discharge can assist primary care physicians in managing these patients. Furthermore, sharing this prediction among multiple healthcare professionals contributes to planning preventive care tailored to individual patients, thereby reducing the risk of readmission in patients with HF.\u003c/p\u003e \u003cp\u003eMachine learning (ML) models developed by integrating multiple factors have been applied to predict outcomes in patients with HF and are expected to outperform conventional statistical methods in predictive accuracy\u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e. Although many studies have demonstrated the utility of ML models in predicting short-term outcomes, such as morbidity, in-hospital mortality, and 30-day rehospitalization rates\u003csup\u003e[\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e, the effectiveness of these models in predicting medium- to long-term prognosis remains unclear. Previous ML models have predominantly relied on pathophysiological factors of the disease, such as clinical laboratory values, electrocardiographic features, echocardiographic findings, and medications\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e. However, these models have not accounted for information recorded by medical professionals other than doctors, including physical conditions, nursing care needs, medication adherence, and the social background of patients\u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e. Comprehensive integration of these variables is crucial for improving medium-term prognostic accuracy.\u003c/p\u003e \u003cp\u003eIn this study, we aimed to develop and validate ML models that incorporate the physical status of patients in addition to clinical laboratory data and treatment details to predict the prognosis of patients with HF within 180 days of discharge. We selected ML models such as conventional logistic regression models with linear feature extraction and tree-based prediction models with nonlinear feature extraction. We evaluated the discrimination of these models by using the area under the receiver operating characteristic curve (AUROC), the area under the precision-recall curve (AUPRC), and the agreement between the models\u0026rsquo; predicted probabilities and observed outcomes by calibration.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Design and Data Collection\u003c/h2\u003e \u003cp\u003eThis multicenter cohort study was conducted using an inpatient database from four affiliated hospitals of Nippon Medical School, including Nippon Medical School Hospital, Musashi Kosugi Hospital, Tama Nagayama Hospital, and Chiba Hokusoh Hospital. The database was constructed using Diagnosis Procedure Combination (DPC) data and laboratory test values from medical records. Data for this study were accessed on July 21, 2024. The authors had no access to information that could identify individual participants during or after data collection. This study was approved by the Central Ethics Review Committee of Nippon Medical School (approval number M-2024-178) and conducted in accordance with the Declaration of Helsinki. Participant consent was obtained using an opt-out method. Reporting followed the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) guidelines\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePatient Selection and Endpoints\u003c/h3\u003e\n\u003cp\u003ePatients with HF in this study included those aged\u0026thinsp;\u0026ge;\u0026thinsp;18 years who were emergently hospitalized in the Department of Cardiovascular Medicine or the Cardiovascular Intensive Care Unit at one of the four Nippon Medical School-affiliated hospitals between April 2018 and September 2023. The eligibility criteria also required a brain natriuretic peptide (BNP) level of \u0026ge;\u0026thinsp;100 pg/mL or an N-terminal pro-BNP (NT-proBNP) level of \u0026ge;\u0026thinsp;300 pg/mL during hospitalization, in accordance with international definitions of HF\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e. Exclusions were made for patients with a hospital stay of \u0026lt;\u0026thinsp;5 days, those discharged owing to death or transfer to another hospital, or those lacking event occurrence with \u0026lt;\u0026thinsp;180 days of follow-up. The endpoint was a composite of all-cause mortality and emergency readmission within 180 days of discharge.\u003c/p\u003e\n\u003ch3\u003eVariables\u003c/h3\u003e\n\u003cp\u003eThe DPC database collected data on all hospitalized patients, including demographics (age, sex), physical metrics height, weight, body mass index (BMI; calculated by dividing body weight [kg] by the square of height [m\u003csup\u003e2\u003c/sup\u003e]), and clinical details (prior emergency hospitalization, comorbidities and procedures during hospitalization, and medications at discharge)\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. The number of prior emergency hospitalizations was defined as those within 180 days after discharge. The medications at discharge recommended in the guidelines for HF and coronary artery disease were defined as those of guideline-directed medical therapy (GDMT) (Supplementary \u003cb\u003eTable S1\u003c/b\u003e)\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\u003eMedications other than those included in GDMT were defined as those not included in the GDMT (ni-GDMT) medications. In this study, we focused on the number of ni-GDMT medications because they are considered to reflect the number of comorbidities, their severity, and polypharmacy. The patient\u0026rsquo;s physical status was evaluated using the nursing care needs score at discharge\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. This score is the sum of the level of assistance required for turning over, transferring, oral hygiene, eating, and changing clothes (0 points: no assistance, 1 point: partial assistance, 2 points: full assistance), whether medical instructions were understood (0 points: yes, 1 point: no), and the presence of risky behavior (0 points: no, 2 points: yes) (Supplementary \u003cb\u003eTable S2\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eBlood test data at admission and discharge were obtained from medical records and defined as the first sampling within 3 days of admission and the last sampling within 14 days before discharge, respectively. After applying the inclusion criteria, BNP was converted to NT-proBNP using the following conversion formula: NT-proBNP\u0026thinsp;=\u0026thinsp;10\u003csup\u003e(1.1 \u0026times; log10[BNP] + 0.57)[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e. Numerical data were scaled using the Standard Scaler.\u003c/p\u003e\n\u003ch3\u003eImputation of Missing Data\u003c/h3\u003e\n\u003cp\u003eMissing data for BMI and blood test findings were imputed using a single imputation implemented in the Python MissForest package\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e. All variables were used for analysis because the missing rate for all variables was \u0026lt;\u0026thinsp;20%, which is the threshold for exclusion.\u003c/p\u003e\n\u003ch3\u003ePredictive Model Development\u003c/h3\u003e\n\u003cp\u003eWe randomly divided the dataset into 80% training and 20% test sets using stratified sampling to preserve the endpoint occurrence rates of the original population. We used the following ML algorithms as predictive models: conventional logistic regression (LR), tree-based random forest (RF), extreme gradient boosting (XGB), and light gradient boosting machine (LGBM)\u003csup\u003e[\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn this study, we used recursive feature elimination with cross-validation (RFECV) of each model to select the optimal subset of features specific to each model. RFECV is a robust method that recursively eliminates less important features and builds a model with the remaining features to identify the optimal subset. Specifically, RFECV was implemented for each model using a 10-fold stratified cross-validation strategy to maximize the AUROC. After selecting the most relevant features, hyperparameter tuning was performed for each model using a grid search with cross-validation to determine the optimal model parameters. This step involved evaluating various combinations of hyperparameters and ultimately selecting the combination that maximized the AUROC for each model (Supplementary \u003cb\u003eTable S3\u003c/b\u003e). As a result of the optimization process (Supplementary \u003cb\u003eTable S4\u003c/b\u003e), the hyperparameter set with the highest AUROC score for each model was identified. The models were calibrated using isotonic regression\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eSHapley Additive exPlanations (SHAP) values were used to explain the output of the ML models\u003csup\u003e[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e. SHAP values quantify the contribution of each feature to the predictions made by the models, allowing for a better understanding of the factors driving the model\u0026rsquo;s decisions.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eModel Validation\u003c/h2\u003e \u003cp\u003eThe bootstrap method was applied to the test set to evaluate the performance of the predictive models. A total of 2,500 bootstrap resamples were used to calculate 95% confidence intervals (CIs) for each performance metric. The discriminatory ability of the model was assessed using the AUROC and AUPRC\u003csup\u003e[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e. For the calibration analysis, the predicted probabilities were divided into ten percentiles, and the mean predicted probability of the outcome and observed probability of the outcome for each bin were plotted. The following two indices were calculated to evaluate the calibration: the calibration slope, which indicates the agreement between the predicted probabilities and observations, and the Brier score, which measures the accuracy of the probability predictions. A calibration slope closer to 1 and a Brier score closer to 0 indicate ideal model performance.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eRisk Classification\u003c/h3\u003e\n\u003cp\u003eWe classified the probabilities predicted by the ML models into three categories for risk stratification: Given that the probability of rehospitalization within 1 year for patients with HF is approximately 30%\u003csup\u003e[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e, the categories were defined as low risk (\u0026lt;\u0026thinsp;0.15), middle risk (\u0026gt;\u0026thinsp;0.15 and \u0026lt;\u0026thinsp;0.30), and high risk (\u0026ge;\u0026thinsp;0.30).\u003c/p\u003e \u003cp\u003eSurvival analysis was performed using the Kaplan\u0026ndash;Meier method, and the log-rank test was used to compare survival distributions between the stratified groups.\u003c/p\u003e\n\u003ch3\u003ePackage for Analysis\u003c/h3\u003e\n\u003cp\u003eAll statistical analyses were performed using Python version 3.9.0 (Python Software Foundation, www.python.org\u003c/a\u003e\u003c/span\u003e\u003cspan address=\"http://www.python.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and R software version 4.2.2 Patched (R Foundation for Statistical Computing, Vienna, Austria). A two-tailed test was performed, and a P-value of \u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eStudy Population and Baseline Characteristics\u003c/h2\u003e \u003cp\u003eAmong the 9,519 patients, we excluded 905 who were discharged because of death, 1,402 who were transferred to other facilities, 466 with a hospital stay\u0026thinsp;\u0026lt;\u0026thinsp;5 days, and 1,842 without events and a follow-up period\u0026thinsp;\u0026lt;\u0026thinsp;180 days. Therefore, 4,904 participants were included in the final analysis, with 3,923 (80%) patients allocated to the training dataset and 981 (20%) allocated to the validation dataset (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows the descriptive statistics of the variables included in the training and test datasets before data processing, as well as the selected features. Outcomes occurred in 1,291 (26.3%) patients, and the features of the training and test datasets were well balanced. For the development of the ML models, out of a total of 61 variables, 28 features were selected for LR, 51 for RF, 52 for XGB, and 36 for LGBM (Supplementary \u003cb\u003eTable S5\u003c/b\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\u003ePatient characteristics\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 \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOverall\u003c/p\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;4,904\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTraining dataset\u003c/p\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;3,923\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTest dataset\u003c/p\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;981\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMissing cases, n (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOutcomes, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1291 (26.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1033 (26.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e258 (26.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDemographics\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\u003eAge (years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e77 (68\u0026ndash;84)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e77 (68\u0026ndash;84)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e77 (68\u0026ndash;83)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale sex, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3142 (64.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2514 (64.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e628 (64.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\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\u003e22.8 (20.4\u0026ndash;25.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22.8 (20.3\u0026ndash;25.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22.8 (20.6\u0026ndash;25.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e128 (2.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHospitalization days (day)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e16 (12\u0026ndash;25)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16 (12\u0026ndash;25)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e17 (12\u0026ndash;25)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNursing care needs score at discharge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 (0\u0026ndash;3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (0\u0026ndash;3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (0\u0026ndash;3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrior emergency hospitalizations, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e807 (16.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e633 (16.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e174 (17.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eComorbidities\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\u003eAcute coronary syndrome: ICD10; I200, I21, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,171 (23.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e944 (24.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e227 (23.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAtrial fibrillation: ICD10; I48, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,595 (32.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,286 (32.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e309 (31.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDyslipidemia: ICD10; E78, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2,621 (53.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,102 (53.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e519 (52.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetes mellitus: ICD10; E10\u0026ndash;E14, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,672 (34.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,346 (34.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e326 (33.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertension: ICD10; I10\u0026ndash;I12, I15, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3,229 (65.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,599 (66.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e630 (64.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIschemic heart disease: ICD10; I201\u0026ndash;209, I22\u0026ndash;25, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,261 (25.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,007 (25.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e254 (25.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedications at discharge\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\u003eBeta-blockers, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3,396 (69.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,710 (69.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e686 (69.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eACE-Is/ARBs and ARNIs, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3,231 (65.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,569 (65.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e662 (67.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAspirin, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,730 (35.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,394 (35.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e336 (34.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eP2Y12 inhibitors, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,731 (35.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,393 (35.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e338 (34.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDirect oral anticoagulants, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,815 (37.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,456 (37.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e359 (36.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWarfarin, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e454 (9.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e352 (9.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e102 (10.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLoop diuretics, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2,772 (56.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,188 (55.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e584 (59.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMineral corticoid receptor antagonists, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,853 (37.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,473 (37.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e380 (38.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTolvaptan, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e983 (20.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e772 (19.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e211 (21.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSGLT2 inhibitors, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e849 (17.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e693 (17.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e156 (15.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStatins, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2,687 (54.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,170 (55.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e517 (52.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProton pump inhibitors, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,766 (36.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,414 (36.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e352 (35.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePimobendan, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e190 (3.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e151 (3.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e39 (4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGout medications, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,551 (31.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,221 (31.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e330 (33.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePsychiatric drugs, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e453 (9.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e369 (9.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e84 (8.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypnotics, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,465 (29.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,164 (29.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e301 (30.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber of ni-GDMT medications, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 (2\u0026ndash;7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (2\u0026ndash;7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4 (2\u0026ndash;7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIn-hospital treatments\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\u003eCatecholamines, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e680 (13.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e550 (14.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e130 (13.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOpioids, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e435 (8.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e352 (9.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e83 (8.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntravenous loop diuretics dosage (mg)/hospitalization\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40 (0\u0026ndash;200)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e40 (0\u0026ndash;200)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e40 (0\u0026ndash;220)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVentilator, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e854 (17.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e678 (17.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e176 (17.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMechanical circulatory support, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e246 (5.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e196 (5.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e50 (5.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHemodialysis, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e227 (4.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e185 (4.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e42 (4.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLaboratory findings at admission\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\u003eWhite blood cells (10\u003csup\u003e3\u003c/sup\u003e/\u0026micro;L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.6 (5.9\u0026ndash;10.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.7 (5.9\u0026ndash;10.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7.4 (5.7\u0026ndash;10.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e20 (0.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHemoglobin (g/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12.4 (10.7\u0026ndash;14.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12.4 (10.6\u0026ndash;14.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12.6 (10.8\u0026ndash;14.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e18 (0.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAspartate transaminase (IU/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29 (21\u0026ndash;46)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29 (21\u0026ndash;45)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e31 (22\u0026ndash;48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e176 (3.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlanine transaminase (IU/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21 (14\u0026ndash;35)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21 (14\u0026ndash;35)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (14\u0026ndash;39)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e148 (3.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlbumin (g/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.7 (3.4\u0026ndash;4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.7 (3.4\u0026ndash;4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.7 (3.4\u0026ndash;4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e76 (1.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlood urea nitrogen (mg/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21.9 (16.2\u0026ndash;31.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21.9 (16.3\u0026ndash;31.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e21.7 (16.0\u0026ndash;32.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e18 (0.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCreatine kinase (IU/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e105 (65\u0026ndash;190)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e106 (65\u0026ndash;192)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e101 (65\u0026ndash;184)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e49 (1.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUric acid (mg/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6.2 (5.0\u0026ndash;7.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6.20 (5\u0026ndash;7.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6.1 (4.9\u0026ndash;7.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e405 (8.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eeGFR (mL/min/1.73 m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e46 (30\u0026ndash;62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e46 (30\u0026ndash;62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e47 (31\u0026ndash;63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e18 (0.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSodium (mEq/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e140 (137\u0026ndash;142)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e140 (137\u0026ndash;142)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e140 (137\u0026ndash;142)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e17 (0.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePotassium (mEq/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4.2 (3.8\u0026ndash;4.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4.2 (3.8\u0026ndash;4.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.2 (3.8\u0026ndash;4.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e18 (0.4)\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\u003e2,713 (995\u0026ndash;6708)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,681 (985\u0026ndash;6724)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2,863 (1021\u0026ndash;6528)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e167 (3.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLaboratory findings at discharge\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\u003eWhite blood cells (10\u003csup\u003e3\u003c/sup\u003e/\u0026micro;L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.8 (4.7\u0026ndash;7.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5.8 (4.7\u0026ndash;7.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5.7 (4.7\u0026ndash;6.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHemoglobin (g/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11.9 (10.4\u0026ndash;13.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11.8 (10.4\u0026ndash;13.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12.0 (10.4\u0026ndash;13.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3 (0.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAspartate transaminase (IU/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22 (17\u0026ndash;28)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22 (17\u0026ndash;28)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (18\u0026ndash;29)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e179 (3.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlanine transaminase (IU/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e18 (12\u0026ndash;27)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18 (12\u0026ndash;27)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e18 (12\u0026ndash;28)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e150 (3.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlbumin (g/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.5 (3.2\u0026ndash;3.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.5 (3.2\u0026ndash;3.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.5 (3.2\u0026ndash;3.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e165 (3.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlood urea nitrogen (mg/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20.2 (15.0\u0026ndash;29.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20.1 (15.0\u0026ndash;29.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e20.2 (15.0\u0026ndash;30.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCreatine kinase (IU/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e45 (31\u0026ndash;68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e46 (31\u0026ndash;68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e45 (31\u0026ndash;68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e96 (2.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUric acid (mg/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6.0 (4.8\u0026ndash;7.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6.1 (4.9\u0026ndash;7.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5.9 (4.7\u0026ndash;7.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e393 (8.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eeGFR (mL/min/1.73 m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e48 (33\u0026ndash;63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e48 (32\u0026ndash;63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e48 (33\u0026ndash;63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSodium (mEq/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e140 (138\u0026ndash;142)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e140 (138\u0026ndash;142)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e140 (137\u0026ndash;142)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePotassium (mEq/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4.3 (4.0\u0026ndash;4.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4.3 (4.0\u0026ndash;4.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.3 (4.0\u0026ndash;4.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e811 (16.5)\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\u003e1,527 (717\u0026ndash;3,530)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,544 (723\u0026ndash;3,544)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1,468 (672\u0026ndash;3,506)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3 (0.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eData are expressed as medians (interquartile ranges). ICD, International Statistical Classification of Diseases and Related Health Problems; ACE-Is, angiotensin-converting enzyme inhibitors; ARBs, angiotensin II receptor blockers; ARNIs, angiotensin receptor neprilysin inhibitors; SGLT2, sodium-glucose cotransporter 2; ni-GDMT, not included in guideline-directed medical therapy; eGFR, estimated glomerular filtration rate; NT-proBNP, N-terminal pro-brain natriuretic peptide\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eEvaluation of Prediction Models\u003c/h2\u003e \u003cp\u003eAfter training the models, the AUROC was calculated based on the results of the predicted events in the test dataset using the trained ML models. The AUROC values for LR, RF, XGB, and LGBM were 0.734 (95% CI, 0.698\u0026ndash;0.769), 0.753 (95% CI, 0.718\u0026ndash;0.786), 0.744 (95% CI, 0.708\u0026ndash;0.778), and 0.752 (95% CI, 0.716\u0026ndash;0.785), respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). Consequently, the models were evaluated by AUPRC, which provides more informative insights into binary classification predictive models with imbalanced data, and the values for LR, RF, XGB, and LGBM were 0.505 (95% CI, 0.445\u0026ndash;0.568), 0.513 (95% CI, 0.454\u0026ndash;0.579), 0.524 (95% CI, 0.463\u0026ndash;0.589), and 0.515 (95% CI, 0.454\u0026ndash;0.582), respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eCalibration was assessed using a plot diagram of predicted probabilities and observations across the ten deciles of predicted risk, and favorable agreement between both values was observed in all models (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC). As calibration indices for the models, values for the calibration slope (LR, 0.964; RF, 1.046; XGB, 1.197; LGBM, 1.088) and Brier score (LR, 0.167; RF, 0.164; XGB, 0.165; LGBM, 0.164) were calculated (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Furthermore, the observed incidence of events was compared across the three defined risk categories (low, middle, and high risk). The observed event rates for each stratified group were within the predicted probability ranges (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Survival analysis using the Kaplan\u0026ndash;Meier method demonstrated that the cumulative observed event rate significantly increased with risk stratification across the four models (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\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\u003eCalibration indices\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCalibration slope (95% confidence interval)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBrier score (95% confidence interval)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLogistic regression\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.964 (0.804\u0026ndash;1.120)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.167 (0.154\u0026ndash;0.181)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRandom forests\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.046 (0.890\u0026ndash;1.260)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.164 (0.151\u0026ndash;1.177)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eExtreme gradient boosting\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.197 (1.013\u0026ndash;1.377)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.165 (0.153\u0026ndash;0.178)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLight gradient boosting machine\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.088 (0.927\u0026ndash;1.250)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.164 (0.152\u0026ndash;0.177)\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 \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\u003eObserved outcomes for each risk stratum classified by predicted probability\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=\"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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRisk stratum\u003c/p\u003e \u003cp\u003e(range of predictive probability)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNumber of patients\u003c/p\u003e \u003cp\u003e(total)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNumber of patients\u003c/p\u003e \u003cp\u003e(outcomes)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eObserved outcome rate\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic regression\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLow: \u0026lt;0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e344\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMiddle: \u0026ge;0.15, \u0026lt;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e241\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHigh: \u0026ge;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e396\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e171\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRandom forests\u003c/b\u003e\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 \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLow: \u0026lt;0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e274\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMiddle: \u0026ge;0.15, \u0026lt;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e350\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHigh: \u0026ge;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e357\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e166\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.46\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eExtreme gradient boosting\u003c/b\u003e\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 \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLow: \u0026lt;0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e262\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMiddle: \u0026ge;0.15, \u0026lt;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e347\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHigh: \u0026ge;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e372\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLight gradient boosting machine\u003c/b\u003e\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 \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLow: \u0026lt;0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e282\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMiddle: \u0026ge;0.15, \u0026lt;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e317\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHigh: \u0026ge;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e382\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e169\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.44\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 \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eModel Interpretation\u003c/h2\u003e \u003cp\u003eThe contribution of each feature to the prediction of outcomes was visualized using the SHAP algorithm. The features were ranked in descending order of importance scores, with red indicating high influence and blue indicating low influence. Established risk factors for HF, such as age, estimated glomerular filtration rate (eGFR), NT-proBNP levels, hemoglobin levels, albumin levels, and the number of previous hospitalizations, were ranked among the top contributors in all four models. Furthermore, nursing care needs scores and the number of ni-GDMT medications, which were the features we focused on in this study, were identified as highly influential factors for outcomes (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). According to the SHAP analysis, the nursing care needs scores contributed 7.72% to the outcome prediction in LR, 4.50% in RF, 4.03% in XGB, and 5.92% in LGBM. The SHAP analysis of LGBM did not reveal clear, strong interactions between the nursing care needs scores and other features in the prediction, compared to the interactions observed between other features (\u003cb\u003eFig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this multicenter retrospective cohort study, we developed ML models to predict all-cause mortality and emergency admission within 180 days of discharge in patients with HF. The models were trained on the training dataset and evaluated on the internal validation dataset. The discriminative power of each model fell within the range of AUROC (0.65\u0026ndash;0.78) reported in previous studies on long-term prognosis prediction of patients with HF\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e. Moreover, the conventional LR model and tree-based models (RF, XGB, and LGBM) demonstrated favorable agreement between predicted and observed values in the calibration, and successful risk stratification was achieved in these models.\u003c/p\u003e \u003cp\u003eDiscrimination refers to the ability of a predictive model to distinguish correctly between positive and negative cases. In this study, the tree-based ML models demonstrated equal or better discriminatory power than the conventional LR model did for both AUROC and AUPRC. This is consistent with the concept that tree-based approaches, which make predictions based on nonlinear relationships, are more suitable than LR, which assumes linearity, especially because some features, such as eGFR and BMI, follow a U-shaped curve\u003csup\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e The ML models in this study demonstrated favorable agreement in calibration, which is an indicator of how closely the model\u0026rsquo;s predicted probability matched the observed outcome. Although this evaluation method with calibration is important for stratifying the risk of individual patients and applying predictive models to clinical practice\u003csup\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e, there is a lack of model evaluation with calibration in a predictive model for HF\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e. In this study, the observed event rates in each stratum fell within the predicted probability ranges (low, middle, and high risk), suggesting that appropriate risk stratification for medium-term prognosis was achieved at the time of hospital discharge. In assessing the utility of a model, improperly structured calibration can be problematic because it leads to either over- or underestimation of risk\u003csup\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e. Among the various algorithms reported as calibration methods for predictive models\u003csup\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e, isotonic regression, which is one of the oldest and most commonly used methods, was applied in this study\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e. This approach ensured reasonable calibration was achieved, as indicated by the calibration plot, and effectively addressed the challenge of over- and underestimation of risk. However, emerging research being conducted on more precise calibration algorithms that utilize neural networks suggests the potential for improvement by applying these methods\u003csup\u003e[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe nursing care needs score and the number of ni-GDMT medications highly influenced the prediction across all models, as shown by the SHAP analysis, in addition to well-known prognostic factors of HF\u003csup\u003e[\u003cspan additionalcitationids=\"CR38\" citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. This indicates that daily physical activity, comorbidity status, and polypharmacy are crucial in the prognosis of patients with HF. However, only the nursing care needs score was available to characterize physical activity in this study, as our retrospectively collected dataset did not include other variables related to the patient\u0026rsquo;s physical condition. Frailty and comorbidity burden are associated with rehospitalization within 6 months of discharge in older patients with HF\u003csup\u003e[\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/sup\u003e. Moreover, based on the recent knowledge of multiple comorbidities or social determinants of health, several physical activity parameters and social factors should be added to the features of ML models\u003csup\u003e[\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eWe constructed the dataset for this study using DPC and clinical laboratory data, which are automatically stored in an electronic format and can be readily exported. The use of DPC and electronic medical records has been standardized across Japanese university hospitals, and several clinical studies have utilized similar databases\u003csup\u003e[\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]\u003c/sup\u003e. Therefore, our results are generalizable and can be easily confirmed using data from other hospitals for external validation. However, our dataset only included inpatients at a university hospital that provides advanced and specialized treatments and did not reflect comprehensive HF care in the community region with hospital\u0026ndash;clinic cooperation. Incorporating clinical data from primary care physicians in clinics and patient information (such as social background, physical activity, and living conditions) recorded by healthcare professionals other than doctors into the database would lead to the construction of a more accurate model for predicting the progression of HF. Building such a comprehensive database with a regional hospital\u0026ndash;clinic\u0026ndash;care team will be the next challenge.\u003c/p\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eLimitations\u003c/h2\u003e \u003cp\u003eThis study has some limitations. First, as a retrospective study, the explanatory variables for the ML models were limited to readily available clinical data with few missing values. Data such as echocardiographic and electrocardiographic findings, vital signs, and social factors were not included in this study. Second, this study was conducted solely in healthcare facilities in Japan; therefore, the findings may not apply to patients in other regions. Future studies should validate the ML models using independent external data from various regions and hospitals. Third, only four ML algorithms were used. Other ML algorithms, such as neural networks and support vector machines, may exhibit better predictive performances. Finally, NT-proBNP levels\u0026thinsp;\u0026ge;\u0026thinsp;300 pg/mL and BNP levels\u0026thinsp;\u0026ge;\u0026thinsp;100 pg/mL were used to diagnose HF. However, data on the conversion between BNP and NT-proBNP suggest that the BNP level equivalent to an NT-proBNP level of 300 pg/mL is lower, at approximately 60\u0026ndash;75 pg/mL\u003csup\u003e[\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]\u003c/sup\u003e. This discrepancy might have led to the inaccurate inclusion of patients with HF. Additionally, because a formula for conversion between BNP and NT-proBNP was used for model development, patients assessed with BNP may not have been accurately evaluated.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusions","content":"\u003cp\u003eFor patients with HF, we developed ML models to predict all-cause mortality and emergency admission in the medium term within 180 days after hospital discharge. This study demonstrates that ML models have favorable discrimination and agreement between predicted probabilities and observations. These ML models can be useful in reducing rehospitalization after discharge through risk stratification in individual patients. Furthermore, the influence of nursing care needs on prediction indicates the importance of multidisciplinary collaboration in HF care.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eT N: Data Curation, Formal analysis, Software, Writing- Original draft preparation. K K: Conceptualization, Methodology, Data Curation, Writing- Original draft preparation. S T: Conceptualization, Data curation, Writing - Review \u0026amp; Editing, Project administration. D H: Conceptualization, Methodology, Writing - Review \u0026amp; Editing. T S: Supervision, Validation, Visualization, Writing - Review \u0026amp; Editing. T T: Validation, Visualization, Writing - Review \u0026amp; Editing. Y K: Conceptualization, Methodology, Resources. T Y: Data curation. M M: Data curation. E K: Data curation. N K: Data curation. A S: Visualization, Data curation. T O: Conceptualization, Supervision. S Y: Supervision. Y K: Supervision. K A: Supervision, Funding acquisition, Project administration. All authors reviewed the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003e The authors extend their sincere thanks to all the people involved in patient care, including emergency staff, technicians, medical engineers, nurses, pharmacists, physicians, and surgeons at Nippon Medical School Hospital, Musashi-Kosugi Hospital, Tama Nagayama Hospital, Chiba Hokusoh Hospital, and Nippon Medical School.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eGBD. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990\u0026ndash;2017: A systematic analysis for the Global Burden of Disease Study 2017. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e392\u003c/b\u003e, 1789\u0026ndash;1858 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYasuda, S., Miyamoto, Y. \u0026amp; Ogawa, H. Current status of cardiovascular medicine in the aging society of Japan. \u003cem\u003eCirculation\u003c/em\u003e \u003cb\u003e138\u003c/b\u003e, 965\u0026ndash;967 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOkura, Y. et al. Impending epidemic: Future projection of heart failure in Japan to the year 2055. \u003cem\u003eCirc. J.\u003c/em\u003e \u003cb\u003e72\u003c/b\u003e, 489\u0026ndash;491 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDi Salvo, T. G. \u0026amp; Stevenson, L. W. Interdisciplinary team-based management of heart failure. \u003cem\u003eDis. Manag Health Outcomes\u003c/em\u003e. \u003cb\u003e11\u003c/b\u003e, 87\u0026ndash;94 (2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBłaziak, M. et al. An artificial intelligence approach to guiding the management of heart failure patients using predictive models: A systematic review. \u003cem\u003eBiomedicines\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 2188 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCroon, P. M. et al. Current state of artificial intelligence-based algorithms for hospital admission prediction in patients with heart failure: A scoping review. \u003cem\u003eEur. Heart J. Digit. Health\u003c/em\u003e. \u003cb\u003e3\u003c/b\u003e, 415\u0026ndash;425 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMedhi, D. et al. Artificial intelligence and its role in diagnosing heart failure: A narrative review. \u003cem\u003eCureus\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e, e59661 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, M. Y. \u0026amp; Son, Y. J. Machine learning-based 30-day readmission prediction models for patients with heart failure: A systematic review. \u003cem\u003eEur. J. Cardiovasc. Nurs.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e, 711\u0026ndash;719 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiyagawa, S. et al. Japan heart failure model \u0026ndash; Derivation and accuracy of survival prediction in Japanese heart failure patients. \u003cem\u003eCirc. Rep.\u003c/em\u003e \u003cb\u003e1\u003c/b\u003e, 29\u0026ndash;34 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShin, S. et al. Machine learning vs. conventional statistical models for predicting heart failure readmission and mortality. \u003cem\u003eESC Heart Fail.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 106\u0026ndash;115 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGeorgiev, K. D., Hvarchanova, N., Georgieva, M. \u0026amp; Kanazirev, B. The role of the clinical pharmacist in the prevention of potential drug interactions in geriatric heart failure patients. \u003cem\u003eInt. J. Clin. Pharm.\u003c/em\u003e \u003cb\u003e41\u003c/b\u003e, 1555\u0026ndash;1561 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCollins, G. S. et al. TRIPOD\u0026thinsp;+\u0026thinsp;AI statement: Updated guidance for reporting clinical prediction models that use regression or machine learning methods. \u003cem\u003eBMJ\u003c/em\u003e \u003cb\u003e385\u003c/b\u003e, q902 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBozkurt, B., Coats, A. \u0026amp; Tsutsui, H. Universal definition and classification of heart failure. \u003cem\u003eJ\u003c/em\u003e. \u003cem\u003eCard\u003c/em\u003e. \u003cem\u003eFail\u003c/em\u003e. 27, S1071-9164(21)00050\u0026thinsp;\u0026ndash;\u0026thinsp;6 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHayashida, K., Murakami, G., Matsuda, S. \u0026amp; Fushimi, K. History and profile of Diagnosis Procedure Combination (DPC): Development of a real data collection system for acute inpatient care in Japan. \u003cem\u003eJ. Epidemiol.\u003c/em\u003e \u003cb\u003e31\u003c/b\u003e, 1\u0026ndash;11 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTsutsui, H. et al. JCS/JHFS 2021 guideline focused update on diagnosis and treatment of acute and chronic heart failure. \u003cem\u003eJ. Card Fail.\u003c/em\u003e \u003cb\u003e27\u003c/b\u003e, 1404\u0026ndash;1444 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNakamura, M. et al. JCS 2020 guideline focused update on antithrombotic therapy in patients with coronary artery disease. \u003cem\u003eCirc. J.\u003c/em\u003e \u003cb\u003e84\u003c/b\u003e, 831\u0026ndash;865 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFitzmaurice, G. M. \u0026amp; Laird, N. M. Multivariate analysis: Discrete variables (Logistic regression) in \u003cem\u003eInternational Encyclopedia of the Social \u0026amp; Behavioral Science\u003c/em\u003e 10221\u0026ndash;10228Elsevier, Amsterdam, (2001).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDong, W. et al. Generative adversarial networks for imputing missing data for big data clinical research. \u003cem\u003eBMC Med. Res. Methodol.\u003c/em\u003e \u003cb\u003e21\u003c/b\u003e, 78 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWaljee, A. K. et al. Comparison of imputation methods for missing laboratory data in medicine. \u003cem\u003eBMJ Open.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e, e002847 (2001).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBreiman, L. \u003cem\u003eMach. Learn.\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e, 5\u0026ndash;32 (2001).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, T., Guestrin, C. \u0026amp; XGBoost A scalable tree boosting system in: \u003cem\u003eProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining\u003c/em\u003e 785\u0026ndash;794ACM, New York, USA, (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMachado, M. R., Karray, S., de Sousa, I. T. \u0026amp; LightGBM An effective decision tree gradient boosting method to predict customer loyalty in the finance industry. In: 14th International Conference on Computer Science \u0026amp; Education (ICCSE); vol 2019. 1111\u0026ndash;1116IEEE Publs, Toronto, (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZadrozny, B. \u0026amp; Elkan, C. Transforming classifier scores into accurate multiclass probability estimates in \u003cem\u003eProceedings of the 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining\u003c/em\u003e; New York 694\u0026ndash;699 (ACM Press; 2002). (2002).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg, S. M. \u0026amp; Lee, S. I. A unified approach to interpreting model predictions in \u003cem\u003eAdv\u003c/em\u003e. \u003cem\u003eNeural Inf\u003c/em\u003e. \u003cem\u003eProcess\u003c/em\u003e. \u003cem\u003eSyst\u003c/em\u003e. 30, 4768\u0026ndash;4777 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarrell, F. E. \u003cem\u003eRegression Modeling Strategies\u003c/em\u003e: \u003cem\u003eWith Applications to Linear Models\u003c/em\u003e, \u003cem\u003eLogistic Regression\u003c/em\u003e, \u003cem\u003eand Survival Analysis\u003c/em\u003eSpringer, New York,. (2001).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaito, T. \u0026amp; Rehmsmeier, M. The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. \u003cem\u003ePLOS One\u003c/em\u003e. \u003cb\u003e10\u003c/b\u003e, e0118432 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIde, T. et al. Clinical characteristics and outcomes of hospitalized patients with heart failure from the large-scale Japanese Registry of Acute Decompensated Heart Failure (JROADHF). \u003cem\u003eCirc. J.\u003c/em\u003e \u003cb\u003e85\u003c/b\u003e, 1438\u0026ndash;1450 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShiraishi, Y. et al. 9-year trend in the management of acute heart failure in Japan: A report from the national consortium of acute heart failure registries. \u003cem\u003eJ. Am. Heart Assoc.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, e008687 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, J. et al. Body mass index and all-cause mortality in heart failure patients with normal and reduced ventricular ejection fraction: A dose-response meta-analysis. \u003cem\u003eClin. Res. Cardiol.\u003c/em\u003e \u003cb\u003e108\u003c/b\u003e, 119\u0026ndash;132 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWatanabe, Y. et al. Fractional excretion of urea nitrogen can identify true worsening renal function in patients with heart failure. \u003cem\u003eESC Heart Fail.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 2043\u0026ndash;2054 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWalsh, C. G., Sharman, K. \u0026amp; Hripcsak, G. Beyond discrimination: A comparison of calibration methods and clinical usefulness of predictive models of readmission risk. \u003cem\u003eJ. Biomed. Inf.\u003c/em\u003e \u003cb\u003e76\u003c/b\u003e, 9\u0026ndash;18 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbdul-Samad, K. et al. Comparison of machine learning and conventional statistical modeling for predicting readmission following acute heart failure hospitalization. \u003cem\u003eAm. Heart J.\u003c/em\u003e \u003cb\u003e277\u003c/b\u003e, 93\u0026ndash;103 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChristodoulou, E. et al. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. \u003cem\u003eJ. Clin. Epidemiol.\u003c/em\u003e \u003cb\u003e110\u003c/b\u003e, 12\u0026ndash;22 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlba, A. C. et al. Discrimination and calibration of clinical prediction models: Users\u0026rsquo; guides to the medical literature. \u003cem\u003eJAMA\u003c/em\u003e \u003cb\u003e318\u003c/b\u003e, 1377\u0026ndash;1384 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, Y., Li, W., Macheret, F. \u0026amp; Gabriel, R. A. Ohno-Machado, L. A tutorial on calibration measurements and calibration models for clinical prediction models. \u003cem\u003eJ. Am. Med. Inf. Assoc.\u003c/em\u003e \u003cb\u003e27\u003c/b\u003e, 621\u0026ndash;633 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo, C., Pleiss, G., Sun, Y. \u0026amp; Weinberger, K. Q. On calibration of modern neural networks in P\u003cem\u003eroceedings of the 34th International Conference on Machine Learning\u003c/em\u003e. Sydney, Australia 1321\u0026ndash;1330. (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchaub, J. A. et al. Amino-terminal pro-B-type natriuretic peptide for diagnosis and prognosis in patients with renal dysfunction: A systematic review and meta-analysis. \u003cem\u003eJACC Heart Fail.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e, 977\u0026ndash;989 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXia, H., Shen, H., Cha, W. \u0026amp; Lu, Q. The prognostic significance of anemia in patients with heart failure: A meta-analysis of studies from the last decade. \u003cem\u003eFront. Cardiovasc. Med.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 632318 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNishikido, T., Oyama, J. I., Nagatomo, D. \u0026amp; Node, K. A reduction of BMI predicts the risk of rehospitalization and cardiac death in non-obese patients with heart failure. \u003cem\u003eInt. J. Cardiol.\u003c/em\u003e \u003cb\u003e276\u003c/b\u003e, 166\u0026ndash;170 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOkoye, C. et al. Predicting mortality and re-hospitalization for heart failure: A machine-learning and cluster analysis on frailty and comorbidity. \u003cem\u003eAging Clin. Exp. Res.\u003c/em\u003e \u003cb\u003e35\u003c/b\u003e, 2919\u0026ndash;2928 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSterling, M. R. et al. Social determinants of health and 90-day mortality after hospitalization for heart failure in the REGARDS study. \u003cem\u003eJ. Am. Heart Assoc.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, e014836 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDickens, C., Dickson, V. V. \u0026amp; Piano, M. R. Perceived stress among patients with heart failure who have low socioeconomic status: A mixed-methods study. \u003cem\u003eJ. Cardiovasc. Nurs.\u003c/em\u003e \u003cb\u003e34\u003c/b\u003e, E1\u0026ndash;E8 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKanazawa, N. et al. Existing data sources for clinical epidemiology: Database of the National Hospital Organization in Japan. \u003cem\u003eClin. Epidemiol.\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 689\u0026ndash;698 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKasahara, S. et al. Conversion formula from B-type natriuretic peptide to N-terminal proBNP values in patients with cardiovascular diseases. \u003cem\u003eInt. J. Cardiol.\u003c/em\u003e \u003cb\u003e280\u003c/b\u003e, 184\u0026ndash;189 (2019).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"mortality, rehospitalization, receiver operating characteristic curve, precision-recall curve, calibration slope, Brier score","lastPublishedDoi":"10.21203/rs.3.rs-6008877/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6008877/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe number of patients with heart failure (HF) is increasing with the aging population, shifting care from hospitals to clinics. Although predicting medium-term prognosis after discharge can enhance care and reduce readmissions, yet no established model has been evaluated for both discrimination and calibration. This multicenter study developed and validated machine learning (ML) models\u0026mdash;including logistic regression, random forests, extreme gradient boosting, and light gradient boosting\u0026mdash; to predict 180-day mortality or emergency hospitalization in 4,904 HF patients with HF. Patients were randomly split into training and validation sets (8:2), and models were trained and evaluated accordingly. All models showed acceptable performance based on the area under the precision-recall curve, good calibration according to the calibration slope and Brier score, and effective risk stratification. The SHapley Additive exPlanations algorithm identified nursing care needs as a key predictor alongside established laboratory values for HF prognosis. ML models effectively predict the 180-day prognosis patients with HF, with nursing care needs highlighting the importance of multidisciplinary collaboration.\u003c/p\u003e \u003cp\u003e \u003cb\u003eClinical Trial Registration\u003c/b\u003e: URL: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.umin.ac.jp/ctr\u003c/span\u003e\u003cspan address=\"https://www.umin.ac.jp/ctr\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e; unique identifier: UMIN000054854\u003c/p\u003e","manuscriptTitle":"Machine learning models for predicting medium-term heart failure prognosis: Discrimination and calibration analyses","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-02-24 06:48:24","doi":"10.21203/rs.3.rs-6008877/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3c1ede74-eefa-4a41-971e-c50b049619bc","owner":[],"postedDate":"February 24th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":44211579,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":44211580,"name":"Health sciences/Cardiology"},{"id":44211581,"name":"Health sciences/Health care"},{"id":44211582,"name":"Health sciences/Health care/Prognosis"}],"tags":[],"updatedAt":"2025-10-08T05:24:19+00:00","versionOfRecord":[],"versionCreatedAt":"2025-02-24 06:48:24","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6008877","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6008877","identity":"rs-6008877","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}