Dear Editor,
Fluid balance management in critically ill heart failure (HF) patients remains a formidable clinical challenge. While clinicians typically aim for net negative fluid balance to alleviate symptoms, recent studies employing fixed strategies have yielded inconsistent results. 1 , 2 The 2024 Heart Failure Association guidelines of the European Society of Cardiology emphasized the importance of individualized fluid balance strategies, particularly for critically ill patients. 3 Our study introduces a novel approach using unsupervised learning to identify four distinct phenotypes of critically ill HF patients, each with unique clinical characteristics and fluid balance requirements. To facilitate clinical application, we have developed a user‐friendly interface that enables rapid phenotype identification and customized fluid management.
We utilized two non‐overlapping databases: III‐CareVue subset and IV versions of the Intensive Care Medical Information Marketplace (MIMIC) 4 for training cohorts and the eICU Collaborative Research Database (eICU) 5 for external validation (Method S1). The MIMIC cohort comprised 5998 patients, while the eICU cohort included 2549 patients (Figure S1). We initially extracted 56 variables from the first day of ICU admission. After eliminating variables with more than 30% missing data, 47 variables remained, encompassing demographics, comorbidities, laboratory values, vital signs, interventions, and severity scores. To ensure a balanced contribution of characteristics, all data underwent cleaning and normalization (Method S2, Figure S2). In‐hospital mortality served as our primary outcome, with ICU length of stay and total hospital length of stay as secondary outcomes.
Uniform Manifold Approximation and Projection (UMAP) was used to determine that there were no differences in clinical characteristics between the two training databases (Figure S3). To classify patients, we applied the K‐prototypes clustering algorithm, which effectively accommodates mixed numerical and categorical attributes while preserving the characteristics of factorial variables (Method S3). The optimal number of clusters was determined using standard tests, considering both statistical metrics and clinical relevance. This approach ultimately identified four distinct phenotypes (Figure S4).
Comparative analysis of these phenotypes revealed distinct clinical profiles (Figure 1, Table 1, Table S1). Phenotype A was characterized by aggressive interventions and inflammation, including high rates of vasoactive drug use, antibiotic use, and mechanical ventilation. This group also exhibited the highest white blood cell count and chloride levels, coupled with the lowest platelet count. Phenotype B represented the mildest form with the most favourable prognosis. Phenotype C was distinguished by the highest mean age, lowest body weight, higher comorbidity burden, and second‐highest mortality rate, despite having the lowest Sequential Organ Failure Assessment (SOFA) score. Phenotype D presented the most severe clinical profile with a poor prognosis. Short‐and long‐term survival outcomes differed significantly among these phenotypes, with Phenotype D showing the worst prognosis and Phenotype B the best prognosis (Figure 2A,B). To validate our findings, we conducted correlation analysis among continuous variables and excluded highly correlated factors before re‐clustering. The resulting phenotypes retained consistent characteristics, confirming the stability of our clustering method (Figures S5–S7).
FIGURE 1.
Characteristics of HF patients with different phenotypes. (A) Histogram of continuous variables for different phenotypes scaled to −1 to 1. (B) Chord diagram of the proportion of comorbidities by phenotype. Arc links between nodes with weights (width of arc). (C) Using Uniform Manifold Approximation and Projection (UMAP) to reduce dimensionality and visualize differences among the four phenotypes. BUN, blood urea nitrogen; DBP, diastolic blood pressure; HR, heart rate; MBP, mean blood pressure; RBC, red blood cell; RDW, red cell distribution width; RR, respiratory rate; SBP, systolic blood pressure; SOFA, Sequential Organ Failure Assessment; SpO2, percutaneous arterial oxygen saturation; WBC, white blood cell.
TABLE 1.
Comparison of baseline characteristics of patients with four phenotypes.
| Heart failure phenotype | |||||
|---|---|---|---|---|---|
| Variables | A (n = 1949) | B (n = 1098) | C (n = 1677) | D (n = 1274) | P‐value 1 |
| Clinical information | |||||
| Admission Age, M (Q₁, Q₃) | 73.42 (65.64,79.90) | 57.46 (49.14,63.53) | 79.73 (74.05,84.17) | 75.73 (68.40,81.90) | <.001*** |
| Gender, n (%) | <.001*** | ||||
| Female | 1077 (55.26) | 340 (30.97) | 765 (45.62) | 417 (32.73) | |
| Male | 872 (44.74) | 758 (69.03) | 912 (54.38) | 857 (67.27) | |
| Race, n (%) | <.001*** | ||||
| White | 1327 (68.09) | 658 (59.93) | 1200 (71.56) | 880 (69.07) | |
| Black | 114 (5.85) | 135 (12.30) | 126 (7.51) | 125 (9.81) | |
| Other | 508 (26.06) | 305 (27.78) | 351 (20.93) | 269 (21.11) | |
| Weight, M (Q₁, Q₃) | 79.50 (66.80,94.25) | 94.78 (77.70,115.39) | 76.80 (65.00,90.25) | 83.18 (70.10,98.15) | <.001*** |
| Admission type, n (%) | <.001*** | ||||
| Emergency | 1398 (71.73) | 918 (83.61) | 1443 (86.05) | 1043 (81.87) | |
| Floor | 551 (28.27) | 180 (16.39) | 234 (13.95) | 231 (18.13) | |
| First care unit, n (%) | <.001*** | ||||
| CICU | 1370 (70.29) | 647 (58.93) | 836 (49.85) | 661 (51.88) | |
| MICU | 411 (21.09) | 322 (29.33) | 514 (30.65) | 458 (35.95) | |
| SICU | 83 (4.26) | 64 (5.83) | 189 (11.27) | 91 (7.14) | |
| NSICU | 5 (.26) | 22 (2.00) | 54 (3.22) | 18 (1.41) | |
| TSICU | 80 (4.10) | 43 (3.92) | 84 (5.01) | 46 (3.61) | |
| Comorbidities | |||||
| Charlson comorbidity index, M (Q₁, Q₃) | 6.00 (5.00,7.00) | 4.00 (3.00,5.00) | 7.00 (6.00,8.00) | 8.00 (7.00,10.00) | <.001*** |
| MI, n (%) | <.001*** | ||||
| No | 1324 (67.93) | 775 (70.58) | 1070 (63.80) | 755 (59.26) | |
| Yes | 625 (32.07) | 323 (29.42) | 607 (36.20) | 519 (40.74) | |
| Hypertension, n (%) | <.001*** | ||||
| No | 719 (36.89) | 466 (42.44) | 478 (28.50) | 215 (16.88) | |
| Yes | 1230 (63.11) | 632 (57.56) | 1199 (71.50) | 1059 (83.12) | |
| Cardiac arrest, n (%) | .015** | ||||
| No | 1876 (96.25) | 1042 (94.90) | 1625 (96.90) | 1210 (94.98) | |
| Yes | 73 (3.75) | 56 (5.10) | 52 (3.10) | 64 (5.02) | |
| AF, n (%) | <.001*** | ||||
| No | 1043 (53.51) | 764 (69.58) | 640 (38.16) | 527 (41.37) | |
| Yes | 906 (46.49) | 334 (30.42) | 1037 (61.84) | 747 (58.63) | |
| Pulmonary disease, n (%) | <.001*** | ||||
| No | 1279 (65.62) | 730 (66.48) | 974 (58.08) | 800 (62.79) | |
| Yes | 670 (34.38) | 368 (33.52) | 703 (41.92) | 474 (37.21) | |
| Renal disease, n (%) | <.001*** | ||||
| No | 1688 (86.61) | 973 (88.62) | 1257 (74.96) | 225 (17.66) | |
| Yes | 261 (13.39) | 125 (11.38) | 420 (25.04) | 1049 (82.34) | |
| Diabetes, n (%) | <.001*** | ||||
| No | 1324 (67.93) | 728 (66.30) | 1119 (66.73) | 430 (33.75) | |
| Yes | 625 (32.07) | 370 (33.70) | 558 (33.27) | 844 (66.25) | |
| Liver disease, n (%) | <.001*** | ||||
| No | 1830 (93.89) | 996 (90.71) | 1609 (95.95) | 1126 (88.38) | |
| Yes | 119 (6.11) | 102 (9.29) | 68 (4.05) | 148 (11.62) | |
| Malignant cancer, n (%) | <.001*** | ||||
| No | 1842 (94.51) | 1057 (96.27) | 1492 (88.97) | 1142 (89.64) | |
| Yes | 107 (5.49) | 41 (3.73) | 185 (11.03) | 132 (10.36) | |
| Rheumatic disease, n (%) | .003*** | ||||
| No | 1867 (95.79) | 1075 (97.91) | 1598 (95.29) | 1229 (96.47) | |
| Yes | 82 (4.21) | 23 (2.09) | 79 (4.71) | 45 (3.53) | |
| Sepsis, n (%) | <.001*** | ||||
| No | 1747 (89.64) | 998 (90.89) | 1564 (93.26) | 1037 (81.40) | |
| Yes | 202 (10.36) | 100 (9.11) | 113 (6.74) | 237 (18.60) | |
| SOFA, M (Q₁, Q₃) | 6.00 (4.00,8.00) | 4.00 (2.00,6.00) | 3.00 (2.00,4.00) | 7.00 (5.00,9.00) | <.001*** |
| Interventions | |||||
| Vasoactive drugs, n (%) | <.001*** | ||||
| No | 461 (23.65) | 755 (68.76) | 1462 (87.18) | 636 (49.92) | |
| Yes | 1488 (76.35) | 343 (31.24) | 215 (12.82) | 638 (50.08) | |
| Antibiotics, n (%) | <.001*** | ||||
| No | 321 (16.47) | 529 (48.18) | 1025 (61.12) | 415 (32.57) | |
| Yes | 1628 (83.53) | 569 (51.82) | 652 (38.88) | 859 (67.43) | |
| Diuretics, n (%) | .014** | ||||
| No | 935 (47.97) | 467 (42.53) | 734 (43.77) | 575 (45.13) | |
| Yes | 1014 (52.03) | 631 (57.47) | 943 (56.23) | 699 (54.87) | |
| CRRT, n (%) | <.001*** | ||||
| No | 1926 (98.82) | 1087 (99.00) | 1669 (99.52) | 1114 (87.44) | |
| Yes | 23 (1.18) | 11 (1.00) | 8 (.48) | 160 (12.56) | |
| Mechanical ventilation, n (%) | <.001*** | ||||
| No | 335 (17.19) | 695 (63.30) | 1351 (80.56) | 796 (62.48) | |
| Yes | 1614 (82.81) | 403 (36.70) | 326 (19.44) | 478 (37.52) | |
| Outcomes | |||||
| Hospital death, n (%) | <.001*** | ||||
| No | 1753 (89.94) | 1024 (93.26) | 1464 (87.30) | 1024 (80.38) | |
| Yes | 196 (10.06) | 74 (6.74) | 213 (12.70) | 250 (19.62) | |
| Los_hospital, M (Q₁, Q₃) | 10.70 (7.19,15.58) | 9.41 (6.11,15.09) | 8.96 (6.08,13.90) | 11.26 (7.43,17.82) | <.001*** |
| Los_icu, M (Q₁, Q₃) | 4.09 (2.88,6.71) | 3.95 (2.83,6.56) | 3.68 (2.65,5.57) | 4.09 (2.83,6.87) | <.001*** |
Note: M: Median, Q₁: 1st Quartile, Q₃: 3st Quartile, * < .05, ** < .01, *** < .001.
Abbreviations: AF, atrial fibrillation; CICU, cardiac intensive care unit; CRRT, continuous renal replacement therapy; Los, length of stay; MI, myocardial infarction; MICU, medical intensive care unit; NSICU, neurosurgical intensive care unit; SICU, surgical intensive care unit; SOFA, sequential organ failure assessment; TSICU, trauma surgical intensive care unit.
Continuous variable was tested using the Kruskal‐waills test. Chi‐square test was used for comparison between categorical variable.
FIGURE 2.
Phenotypic survival curves versus fluid equilibrium strategies, and performance of machine learning models Kaplan–Meier survival analyses at 28 days (A) and 1 year (B) for four phenotypes of HF. Association between daily fluid management strategies and in‐hospital mortality in four phenotypes (C). Feature selection (D) and ROC curve (E) for classifier learners. BUN, blood urea nitrogen; CRRT, continuous renal replacement therapy; DBP, diastolic blood pressure; HR, heart rate; JMIM, Joint Mutual Information Maximiza; MBP, mean blood pressure; RBC, red blood cell; RDW, red cell distribution width; ROC, receiver operating characteristic curve; RR, respiratory rate; SBP, systolic blood pressure; SpO2, percutaneous arterial oxygen saturation; WBC, white blood cell.
Additionally, we investigated phenotype‐specific fluid management strategies using net daily fluid balance data, adjusted for demographic factors, laboratory parameters, vital signs, and interventions. The impact of these strategies on in‐hospital mortality was analyzed using the parametric g‐formula (Method S4), focusing on the first seven days of ICU admission. Figure 2C illustrates how different fluid management strategies influence in‐hospital mortality across the four patient phenotypes over 7 days: Phenotype A presented with severe respiratory failure, shock, and inflammation, and benefited from fluid balances between −1000 and 500 mL daily. This aligned with recent studies on acute respiratory distress syndrome (ARDS) and ventilator‐related events, confirming the adverse effects of positive fluid balance on mechanical ventilation duration and mortality. 6 , 7 Phenotype C, despite milder clinical parameters, exhibited significantly higher mortality rates compared to Phenotype B, likely due to older age and multiple comorbidities. We recommended a daily net fluid balance ranging from −1500 and 500 mL for this group, underscoring the impact of age and frailty on HF prognosis. 8 Phenotype D, characterized by severe metabolic derangements including acidosis, renal dysfunction, and sepsis, had the poorest prognosis. Our results suggested a more restrictive fluid strategy (−2000 to −500 mL daily), consistent with studies demonstrating adverse effects of positive fluid balance in populations with kidney disease and sepsis. 9 , 10 Phenotype B showed no clear benefit from specific fluid management strategies, warranting further investigation to determine whether this reflected the relative mildness of their condition or methodological limitations.
To facilitate the efficient classification of HF phenotypes across different cohorts, we developed a machine learning‐based classification model. The Joint Mutual Information Maximiza (JMIM) method identified nine variables with a feature importance score >.8, including age, blood urea nitrogen (BUN), hematocrit, vasoactive drugs, renal disease, creatinine, diastolic blood pressure (DBP), mechanical ventilation, and anion gap (Figure 2D). Based on benchmark tests (Table S3), we selected the Extreme Gradient Boosting (XGBoost) model for phenotype classification. The model achieved high predictive performance in MIMIC (AUC:.918–.943) (Figure 2E) and satisfactory performance in the eICU cohort used for external validation (AUC:.802–.907) (Figure S9). We evaluated the performance metrics and decision curve analysis of the model, and the results showed that the XGBoost model had good performance and clinical net benefit (Table S4, Figure S8). Then, we conducted an interpretability analysis to visualize the model's decision‐making process (Figures S10–S12). To support clinical application, we developed a web‐based tool (https://7kdtqk‐guanchengcheng.shinyapps.io/hf_phenotype/).
While our study provided valuable insights, several limitations warranted acknowledgement. First, the retrospective nature of the study precluded clear causal inference. Additionally, the absence of some important variables (such as ejection fraction and natriuretic peptides) might have led to the omission of potentially significant factors. Future randomized controlled trials were necessary to confirm the efficacy of phenotype‐specific fluid management strategies. Second, our analysis primarily focused on net fluid intake and its impact on prognosis. Further research was needed to explore the effects of infusion rates and individual fluid responsiveness on patient outcomes.
In summary, we have identified four distinct phenotypes of critically ill heart failure patients, each with unique clinical characteristics and fluid management needs. Our novel classification model and user interface facilitated rapid phenotype identification, enabling personalized fluid management strategies.
AUTHOR CONTRIBUTIONS
Bing Xiao contributed to the research design. Chengjian Guan, Angwei Gong, and Yan Zhao contributed to data collection, data processing, and graphing. Chengjian Guan, Hangtian Yu, and Shuaidan Zhang conducted model construction and deployment. Zhiyi Xie, Yehui Jin, Xiuchun Yang, and Jingchao Lu contributed to data proofreading and formal analysis. Chengjian Guan and Angwei Gong contributed to the writing of the manuscript. Xiao Bing contributed to the review and editing. All authors have read and approved the final manuscript.
CONFLICT OF INTEREST STATEMENT
The authors declare no conflict of interest.
ETHICS STATEMENT
This study was conducted in accordance with the Declaration of Helsinki. Since the public databases used in this study all use de‐identified data, individual informed consent is not required, so we informed the Ethics Committee of this without a written report.
Supporting information
Supporting information
ACKNOWLEDGEMENTS
The authors thank all the administrators who collected, organized, and maintained the MIMIC database and eICU database. The project was supported by the S&T Program of Hebei No. 22377728D.
Chengjian Guan and Angwei Gong contributed equally to this work.
DATA AVAILABILITY STATEMENT
The data used in this study are from the MIMIC and the eICU databases. Although the database is de‐identified, the investigator still needs to obtain permission through the corresponding examination, so we have no right to disclose the data. The investigator may request data from the corresponding author of this article upon reasonable request after completing the examination and obtaining appropriate access.
REFERENCES
- 1. Machado d'Almeida KS, Rabelo‐Silva ER, Souza GC, et al. Aggressive fluid and sodium restriction in decompensated heart failure with preserved ejection fraction: results from a randomized clinical trial. Nutrition. 2018;54:111‐117. [DOI] [PubMed] [Google Scholar]
- 2. Holst M, Strömberg A, Lindholm M, Willenheimer R. Liberal versus restricted fluid prescription in stabilised patients with chronic heart failure: result of a randomised cross‐over study of the effects on health‐related quality of life, physical capacity, thirst and morbidity. Scand Cardiovasc J. 2008;42(5):316‐322. [DOI] [PubMed] [Google Scholar]
- 3. Mullens W, Damman K, Dhont S, et al. Dietary sodium and fluid intake in heart failure. A clinical consensus statement of the Heart Failure Association of the ESC. Eur J Heart Fail. 2024;26(4):730‐741. [DOI] [PubMed] [Google Scholar]
- 4. Johnson AEW, Bulgarelli L, Shen L, et al. MIMIC‐IV, a freely accessible electronic health record dataset. Sci Data. 2023;10(1):1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Pollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eICU Collaborative Research Database, a freely available multi‐center database for critical care research. Sci Data. 2018;5:180178. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Fot EV, Khromacheva NO, Ushakov AA, Smetkin AA, Kuzkov VV, Kirov MY. Optimizing fluid management guided by volumetric parameters in patients with sepsis and ARDS. Int J Mol Sci. 2023;24(10):8768. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Wang W, Zhu S, He Q, et al. Fluid balance and ventilator‐associated events among patients admitted to ICUs in China: a nested case‐control study. Crit Care Med. 2022;50(2):307‐316. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Matsue Y, Kamiya K, Saito H, et al. Prevalence and prognostic impact of the coexistence of multiple frailty domains in elderly patients with heart failure: the FRAGILE‐HF cohort study. Eur J Heart Fail. 2020;22(11):2112‐2119. [DOI] [PubMed] [Google Scholar]
- 9. Beagle AJ, Prasad PA, Hubbard CC, et al. Associations between volume of early intravenous fluid and hospital outcomes in septic patients with and without heart failure: a retrospective cohort study. Crit Care Explor. 2024;6(5):e1082. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Balakumar V, Murugan R, Sileanu FE, Palevsky. Both positive and negative fluid balance may be associated with reduced long‐term survival in the critically ill. Crit Care Med. 2017;45(8):e749‐e757. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supporting information
Data Availability Statement
The data used in this study are from the MIMIC and the eICU databases. Although the database is de‐identified, the investigator still needs to obtain permission through the corresponding examination, so we have no right to disclose the data. The investigator may request data from the corresponding author of this article upon reasonable request after completing the examination and obtaining appropriate access.