Phenotyping of Patients With Clinical Features of Advanced Heart Failure: A Latent Class Analysis of a Multicenter Registry of Patients Hospitalized With Heart Failure Authors: Keiichi Izumi, Ryo Nakamaru, Takashi Kohno, Ayumi Goda, Yasuyuki Shiraishi, Shinsuke Takeuchi, Mitsunobu Kitamura, Yuji Nagatomo, Atsushi Mizuno, Michiru Nomoto, Munehisa Sakamoto, Kazutaka Miyamoto, Kyoko Soejima, Shun Kohsaka, Tsutomu Yoshikawa Published: 2026 ABSTRACT AbstractBackgroundAlthough the “I NEED HELP” criteria facilitate timely consideration of advanced heart failure (HF) therapies, patient classification based on the variables included in the criteria are scarcely investigated. This study aimed to identify phenotypically distinct subgroups of patients with advanced HF using unsupervised cluster analysis based on the variables included in these criteria.MethodsA multicenter registry was used to identify patients hospitalized with HF who met at least 1 I NEED HELP criterion. Latent class analysis was performed on 9 variables from the criteria.ResultsA total of 2520 patients (mean age, 74±13 years; 40% women) were included in this study. Latent class analysis identified 3 phenotypes: more de novo HF with the lowest number of applicable criteria (phenotype 1, 31%); younger age with the lowest left ventricular ejection fraction and highest number of applicable criteria (phenotype 2, 19%); and older age with more comorbidities and poorer renal function (phenotype 3, 49%). After multivariable adjustment, phenotypes 2 and 3 were associated with a higher incidence of composite all‐cause death or rehospitalization for HF (phenotype 2: hazard ratio [HR], 1.96 [95% CI, 1.58–2.42]; phenotype 3: HR, 1.54 [95% CI, 1.28–1.84]; phenotype 1: reference) during the follow‐up period (median, 461 [interquartile range, 130–730] days).ConclusionsThis study describes a novel phenotypical classification based on the I NEED HELP criteria, with significantly different prognoses identified for patients with advanced HF. These classifications may facilitate risk stratification in this population with a complicated clinical profile.Keywords: classification, heart failure, I NEED HELP, latent class analysis, prognosisSubject Categories: Heart Failure Nonstandard Abbreviations and Acronyms guideline‐directed medical therapy Assessment of the I Need Help Markers in Heart Failure latent class analysis mineralocorticoid receptor antagonist West Tokyo Heart Failure‐1 Clinical Perspective What Is New? This study identifies a novel phenotypic classification based on the “I NEED HELP” criteria, defining 3 clinically distinct phenotypes: robust, poor cardiac condition, and multimorbidity. This study identifies a novel phenotypic classification based on the “I NEED HELP” criteria, defining 3 clinically distinct phenotypes: robust, poor cardiac condition, and multimorbidity. These phenotypes demonstrate clearly divergent prognoses, with the poor cardiac condition and multimorbidity groups showing significantly higher risks of heart failure rehospitalization and death. These phenotypes demonstrate clearly divergent prognoses, with the poor cardiac condition and multimorbidity groups showing significantly higher risks of heart failure rehospitalization and death. What Are the Clinical Implications? The I NEED HELP framework may enable phenotype‐based risk stratification and support more tailored therapeutic strategies in advanced heart failure. The I NEED HELP framework may enable phenotype‐based risk stratification and support more tailored therapeutic strategies in advanced heart failure. Advanced heart failure (HF) comprises ≈1%–10% of the overall HF population. 1 Patients with advanced HF have a markedly poor prognosis, 2 with annual mortality rates reaching as high as 50% 3 , 4 even with guideline‐based therapies. 5 As ≈5% of patients with HF progress to an advanced state each year, 6 current clinical guidelines 7 , 8 emphasize the significance of early recognition of this condition to facilitate the optimization of care. 1 , 5 To enable the timely identification of patients with advanced HF and promote the prompt consideration of appropriate therapies, the “I NEED HELP” mnemonic has been proposed. 1 I NEED HELP is a mnemonic that integrates 9 elements, including clinical history, prior hospitalizations, medication intolerance, symptoms, and end‐organ dysfunction. The variables included in the mnemonic can assist in the early identification and prognostic stratification of patients with advanced HF across various clinical settings. 7 , 8 , 9 This mnemonic framework would appear to enable the identification of high‐risk patients proceeding to advanced HF status across a broad spectrum of clinical presentations. However, prior studies have evaluated either the number of I NEED HELP criteria met or each criterion in isolation in relation to outcomes, without accounting for the complex interplay of these features. 10 , 11 , 12 Given the substantial heterogeneity in clinical profiles among patients with advanced HF, a more refined framework for patient classification is needed. In particular, understanding how combinations of I NEED HELP criteria manifest in distinct clinical subgroups may provide valuable insight into underlying pathophysiology and assist the development of more individualized therapeutic strategies. Latent class analysis (LCA) is a statistical procedure used to qualitatively identify different subgroups within populations using observed variables. LCA identifies unobserved heterogeneity by clustering individuals on the basis of their responses to observed variables to uncover the underlying patterns or structures in the data. 13 , 14 The LCA approach has been used in several previous studies investigating HF with preserved ejection fraction and heterogeneous pathophysiologies. 15 To the best of our knowledge, no study has used LCA to predict mortality risk in patients with advanced HF. We hypothesized that phenomapping focused on incorporating I NEED HELP criteria can identify subgroups among patients with advanced HF with distinct clinical features and prognosis. We aimed to identify phenotypically distinct subgroups of patients with advanced HF defined by the I NEED HELP criteria using unsupervised clustering analysis. METHODS Data, Analytic Methods (Code), and Research Materials The data supporting the findings of this study are available from the corresponding author upon reasonable request. Study Design and Participants The details of the WET‐HF1 (West Tokyo Heart Failure‐1) registry have been described previously. 14 , 16 In brief, this prospective multicenter cohort registry was designed to collect data on the clinical characteristics and outcomes of patients hospitalized with acute HF who met the Framingham criteria 14 as the primary cause of admission. This registry study was conducted across 3 university hospitals and 3 tertiary referral hospitals in the Tokyo metropolitan area. To ensure a thorough evaluation of patient care and outcomes, data were collected by dedicated clinical research coordinators through medical records and interviews with the treating physicians. Data were entered into an electronic system for data capture, which featured a comprehensive query engine and validation checks to maintain data quality. Outliers in the continuous variables or unexpected values in the categorical variables were flagged according to predefined criteria, and the relevant institution was contacted to verify these discrepancies. Additionally, the quality of reporting was reviewed annually by the principal investigators (Y.S. and S.K.), with periodic queries conducted to ensure data integrity. Patients who declined participation or presented with concurrent HF and acute coronary syndrome were excluded from this study. The study protocol was approved by the institutional review board at each center and was conducted in accordance with the Declaration of Helsinki. Informed consent, either written or oral, was obtained from each participant before their enrollment in the study. Before launching this registry, information about the study's objective, social significance, and a summary were submitted for clinical trial registration to the University Hospital Medical Information Network of Japan (UMIN000001171). Study Population and Assessment of I NEED HELP Criteria We extracted data from 4000 consecutive patients with acute HF in the WET‐HF1 registry between January 2006 and December 2017. After excluding patients who died during the index hospitalization (n=164 [4.1%]), those without follow‐up data (n=244 [6.1%]), those without data on the I NEED HELP criteria (n=683 [17.1%]), and those who did not fulfill the I NEED HELP criteria (n=389 [9.7%]), the remaining participants (n=2520) were included in the current study (Figure S1). To align each criterion of the I NEED HELP framework 9 with variables in the WET‐HF1 registry, we prespecified operational definitions for each component on the basis of available registry data, including pragmatic substitutions where direct equivalents were lacking. The detailed mapping of each criterion is provided in Table S1, with key definitions summarized below: I: Previous or ongoing need of inotropes: the use of dobutamine, dopamine, noradrenaline, adrenaline, or phosphodiesterase III inhibitors during hospitalization. N: Persisting New York Heart Association functional classification III or IV/persistently high natriuretic peptide levels: New York Heart Association class III or IV at discharge. E: End‐organ dysfunction: estimated glomerular filtration rate <30 mL/min per 1.73 m2 at discharge. 17 E: Ejection fraction: left ventricular ejection fraction (LVEF) <20% after HF signs and symptoms are stabilized. D: Recurrent appropriate defibrillator shocks: history of ventricular tachycardia or ventricular fibrillation. H: History of hospitalization: >1 hospitalization with HF before the index hospitalization. E: Edema or escalating diuretics: a dose of furosemide, azosemide or torasemide more than the fourth quartile (furosemide ≥40 mg, azosemide ≥60 mg, torasemide ≥8 mg) in cases in which only loop diuretics are used, 18 use of both loop diuretics and tolvaptan, 19 or use of thiazide and other diuretics 20 at discharge. L: Low blood pressure: systolic blood pressure <100 mm Hg at discharge. P: Inability to up‐titrate or need to reduce/discontinue prognostic medications: the discontinuation or deescalation of angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers, β blockers, or mineralocorticoid receptor antagonists (MRAs). 7 , 8 The doses of β blockers were converted to carvedilol equivalents on the basis of the following equation: 2.5 mg bisoprolol=10 mg carvedilol. 21 Data on the doses of angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers and MRAs were not collected; therefore, these deescalations were not evaluated. Guideline‐Directed Medical Therapy The implementation of guideline‐directed medical therapy (GDMT) was defined as the use of angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers, β blockers, and MRAs 7 , 8 at the time of discharge. Sodium–glucose cotransporter‐2 inhibitors and angiotensin receptor–neprilysin inhibitors were not included in the GDMT, as they were not approved for use in Japan at the time of the study. The eligibility of GDMT was based on the following criteria: LVEF ≤40%, estimated glomerular filtration rate ≥30 mL/min per 1.73 m2, serum potassium level ≤5.2 mmol/L, and heart rate at the time of discharge ≥40 beats/min. 22 Clinical Outcomes The primary outcome was a composite of all‐cause death and hospitalization for HF within 2 years after discharge. The secondary outcomes were all‐cause death, hospitalization for HF, cardiac death, and noncardiac death within the 2‐year period. HF‐related, sudden cardiac, and other cardiac deaths (eg, acute coronary syndrome) were considered cardiac deaths. Noncardiac deaths included all other causes of death. Deaths of unknown cause were not classified as either cardiac or noncardiac deaths and were therefore not counted in cause‐specific mortality analyses. In accordance with standard care practices, the treating physicians at each participating hospital made decisions about rehospitalization for HF. To ensure the accuracy of each end point, the WET‐HF1 registry was supported by a central study committee that adjudicated the end point determination. Statistical Analysis All analyses were conducted using a complete‐case approach for the I NEED HELP variables. First, the prevalence of each I NEED HELP criterion was assessed in the study population. A network plot was constructed to identify the relationships between each criterion and assessed using Cramer's V coefficient. We performed LCA, a statistical approach of unsupervised clustering, 14 , 15 to identify phenotypes based on patterns of the clinical features of advanced HF, incorporating the 9 variables of the I NEED HELP criteria. Maximum likelihood estimates were used to identify latent classes on the basis of deficit patterns, considering groups ranging from 2 to 9. Each participant was assigned to the group with the highest probability of membership. The first minimum value of the Bayesian information criterion determined the optimal number of classes. 23 To evaluate the certainty of class assignment, posterior class membership probabilities were calculated for each individual. For each class, we summarized the mean posterior probability among individuals assigned to that class and examined the distribution of probabilities across classes. A mean posterior probability ≥0.70 was considered indicative of acceptable classification accuracy. 13 , 24 For internal validation, we performed 1000 bootstrap resamples of the original data set. In each resampling data, LCA was performed using the 9 variables from the I NEED HELP criteria. The optimal number of classes was reassessed by comparing the Bayesian information criterion values across bootstrap iterations. We calculated the Jaccard index to evaluate the stability of class assignments; this approach quantified the similarity between class memberships derived from the bootstrap samples and those from the original data set. Additionally, internal validation was performed by randomly splitting the data set into training (70%) and test (30%) subsets. LCA was conducted separately in each subset to assess the reproducibility of the identified class structure. The characteristics and outcomes of the derived classes were then evaluated in both subsets. Patient characteristics were compared among the 3 phenotypes on the basis of the LCA. Parametric and nonparametric variables, along with their respective differences, were assessed using 1‐way ANOVA or the Kruskal–Wallis test. Significant differences between independent categorical variables were assessed using the χ2 test. The proportion of GDMT‐eligible patients within each phenotype and their GDMT prescription rates at discharge were evaluated. The incidence of composite events, all‐cause death, hospitalization for HF, cardiac death, and noncardiac death was estimated using the Kaplan–Meier survival function and compared among the 3 phenotypes using the log‐rank test. Cox proportional hazard analyses were performed to assess the risk of composite end points, all‐cause death, hospitalization for HF, cardiac death, and noncardiac death among the phenotypes. The covariates included in the multivariable model were age, sex, pathogenesis of HF (ischemic cardiomyopathy or not), and atrial fibrillation, as well as body mass index, serum sodium level, and hemoglobin levels at discharge. Components of the I NEED HELP criteria (eg, blood pressure, estimated glomerular filtration rate, and New York Heart Association functional class) were not included as covariates, as adjusting for these variables would compromise the conceptual integrity of the phenotyping approach and undermine the interpretability of phenotype–outcome associations. Natriuretic peptides were not included because of a substantial proportion of missing data. Accordingly, only covariates with low missingness (<5% for most variables and 5%–10% for body mass index) were included in the final models. Statistical significance was defined as P<0.05. All statistical analyses were conducted using R software version 4.4.0 (R Foundation for Statistical Computing, Vienna, Austria). LCA was performed using the “poLCA” package, and bootstrap resampling was conducted using the “foreach” and “doParallel” package in R software. RESULTS Baseline Patient Characteristics A total of 2520 (63.0%) patients from the WET‐HF1 registry who met at least 1 I NEED HELP criterion were analyzed in this study (Figure S1). The baseline characteristics of the study population are detailed in Table S2. Overall, the mean age was 74.0±13.3 years, and 1007 (40.0%) were women; the average LVEF was 44.0%±15.4%. As depicted in Figure 1A, the 3 most common I NEED HELP criteria in the study population were edema/escalating diuretics (62.3%), hospitalization for HF (33.3%), and low blood pressure (26.7%). Approximately 4% of the participants met ≥5 I NEED HELP criteria (Figure 1B). Distribution of the I NEED HELP criteria. The figure depicts the proportion of patients fulfilling each of the 9 I NEED HELP criteria in the study population (A) and different cumulative numbers of the I NEED HELP criteria (from 1 to 9) in the study population (B). A network plot representing the relationships among the criteria is displayed in Figure 2. The most significant association was between the criterion ‘LVEF <20% after HF signs and symptoms stabilized’ and ‘previous or ongoing need of inotropes’ (V=0.146, P<0.001), as well as between ‘inability to up‐titrate or need to reduce/discontinue prognostic medications’ and ‘edema or escalating diuretics’ (V=0.146, P<0.001). Network plot for the I NEED HELP criteria in this study. The circles represent the number of participants. The thickness of each line represents the association between the 2 criteria according to Cramer's V coefficient. NYHA indicates New York Heart Association. Classification of Patients With Advanced HF The optimal number of classified phenotypes based on the current LCA was 3 (Table S3). Comparisons of patient characteristics among the 3 phenotypes are summarized in Table 1 and Table S4. The phenotypic characteristics based on the prevalence of each I NEED HELP criterion are shown in Figure 3 and Figure S2. Table 1Baseline Patient Characteristics According to the PhenotypeVariablesPhenotype 1Phenotype 2Phenotype 3 P valueRobustPoor cardiac conditionMultimorbidityn=787n=490n=1243DemographicsFemale sex314 (39.9)167 (34.1)526 (42.3)0.007Age, y75.0 (64.5–83.0)72.0 (60.3–80.0)79.0 (71.0–85.0)<0.001Body mass index on admission, kg/m2 24.2±4.822.9±4.223.3±4.3<0.001Body mass index at discharge, kg/m2 22.2±4.121.0±3.921.6±4.0<0.001Length of stay during index hospitalization, d14 (10–21)18 (11–28)15 (10–23)<0.001Number of I NEED HELP criteria1 (1–2)3 (2–4)2 (1–3)<0.001Clinical profile at dischargeSystolic blood pressure, mm Hg110.0 (100.0–120.0)92.0 (88.0–98.0)116.0 (106.0–128.0)<0.001Diastolic blood pressure, mm Hg62.5±10.755.6±8.162.7±11.3<0.001Heart rate, bpm71 (64–80)72 (62–80)70 (60–78)0.008New York Heart Association class III–IV153 (19.4)112 (22.9)355 (28.6)<0.001Medical historyHypertension527 (67.0)231 (47.1)909 (73.1)<0.001Dyslipidemia293 (37.4)172 (35.7)508 (41.0)0.073Diabetes223 (28.4)138 (28.2)495 (39.8)<0.001Atrial fibrillation/atrial flutter386 (49.0)269 (54.9)601 (48.4)0.043Ventricular tachycardia/ventricular fibrillation12 (1.5)75 (15.3)17 (1.4)<0.001Chronic obstructive pulmonary disease45 (5.7)22 (4.5)61 (4.9)0.594Previous HF admission0 (0)243 (49.6)595 (47.9)<0.001Stroke or transient ischemic attack96 (12.2)64 (13.1)175 (14.1)0.457Hemodialysis1 (0.1)6 (1.2)79 (6.4)<0.001Pathogenesis of HFIschemic203 (25.8)132 (26.9)406 (32.7)0.002Valvular200 (25.4)102 (20.8)388 (31.2)<0.001Dilated119 (15.1)136 (27.8)113 (9.1)<0.001Laboratory data at dischargeHemoglobin, g/dL12.9±2.112.5±2.111.3±2.0<0.001Sodium, mmol/L138.9±3.3138.0±3.7138.4±3.7<0.001Potassium, mmol/L4.3±0.54.4±0.54.3±0.50.010Estimated glomerular filtration rate, mL/min per 1.73 m2 57.1 (47.3–69.2)55.4 (42.9–69.5)36.7 (22.9–55.6)<0.001Total bilirubin, mg/dL0.7 (0.5–1.0)0.8 (0.6–1.2)0.6 (0.4–0.9)<0.001B‐type natriuretic peptide, pg/mL222 (124–437)317 (169–628)292 (142–570)<0.001N‐terminal pro‐B‐type natriuretic peptide, pg/mL1541 (662–2866)2316 (1233–4571)2505 (1209–6368)<0.001Echocardiographic dataLVEF, %45 (31–57)33 (21–48)48 (35–60)<0.001Medication and device use at discharge, n (%)Loop diuretics787 (100.0)371 (75.7)873 (70.2)<0.001Furosemide equivalent dose, mg20 (20–40)20 (10–40)20 (0–40)<0.001Thiazides48 (6.1)42 (8.7)114 (9.4)0.034Tolvaptan33 (4.5)46 (10.1)72 (6.1)0.001Renin–angiotensin system inhibitors592 (75.2)309 (63.1)678 (54.5)<0.001β blockers617 (78.4)393 (80.2)902 (72.6)<0.001MRAs300 (38.1)231 (47.1)337 (27.1)<0.001Carvedilol equivalent dose, mg5.0 (2.5–10.0)5.0 (2.5–10.0)5.0 (2.5–10.0)0.466Digitalis67 (8.5)54 (11.0)60 (4.8)<0.001Implantable cardioverter‐defibrillator11 (1.4)51 (10.4)34 (2.7)<0.001Cardiac resynchronization therapy0 (0)27 (5.5)18 (1.4)<0.001Values are expressed as mean±SD, median (interquartile range), or number (%).HF indicates heart failure; LVEF, left ventricular ejection fraction; and MRA, mineralocorticoid receptor antagonist. Baseline Patient Characteristics According to the Phenotype Values are expressed as mean±SD, median (interquartile range), or number (%). HF indicates heart failure; LVEF, left ventricular ejection fraction; and MRA, mineralocorticoid receptor antagonist. Heat map of the phenotypic characteristics across the 3 phenotypes. NYHA indicates New York Heart Association. Based on the clinical characteristics, the phenotypes were labeled as robust (phenotype 1: n=787 [31.2%]), poor cardiac condition (phenotype 2: n=490 [19.4%]), or multimorbidity (phenotype 3: n=1243 [49.3%]). Patients in phenotype 1 had the lowest number of applicable I NEED HELP criteria and better preservation of renal function; all patients in this group presented with de novo HF. The proportion of patients meeting the edema or escalating diuretics criterion was the highest in phenotype 1 (Figure 3 and Figure S2). Patients with phenotype 2 had the youngest age, lowest systolic blood pressure, lowest LVEF, and the highest number of applicable criteria. The proportion of dilated cardiomyopathy was higher than that of other phenotypes. Patients with phenotype 3 were older in age, with more comorbidities and poorer renal function. The proportion of patients with ischemic cardiomyopathy and valvular heart disease was the greatest for phenotype 3. Figure S3 graphically presents the posterior probabilities for the 3‐class phenotypes. Each dot represents a participant's probability of belonging to a given phenotype, with colors indicating the assigned class. The posterior class membership probabilities are summarized in Table S5. The mean posterior probabilities for the assigned classes were 0.763 for the robust class, 0.700 for the poor cardiac condition class, and 0.708 for the multimorbidity class, with values meeting or closely approximating the prespecified threshold for acceptable classification. Guideline‐Directed Medical Therapy The proportion of patients meeting the eligibility for GDMT (n=839 [33.3%]) was lower in the multimorbidity group than in the other groups (robust, 41.7%; poor cardiac condition, 54.9%; multimorbidity, 19.5%; P<0.001; Figure 4). Furthermore, among GDMT‐eligible patients, the multimorbidity group had lower prescription rates of renin–angiotensin system inhibitors (robust, 80.2%; poor cardiac condition, 70.3%; multimorbidity, 63.2%; P<0.001) and MRAs (47.6% versus 56.1% versus 43.8%, respectively; P=0.015) than the other phenotypes. In contrast, among GDMT‐eligible patients, no significant difference was observed in the prescription rate of β blockers across the 3 phenotypes (90.2% versus 88.4% versus 86.4%, respectively; P=0.355). Proportion of patients meeting GDMT eligibility in each phenotype. GDMT indicates guideline‐directed medical therapy. Clinical Outcomes The median follow‐up period was 461 (130–730) days. The poor cardiac condition and multimorbidity groups had higher incidences of composite events, hospitalization for HF, all‐cause death, cardiac death, and noncardiac death (Figure 5A–5E). The higher risk of composite events, hospitalization for HF, all‐cause death, and cardiac death in the poor cardiac condition and multimorbidity groups persisted even after multivariable adjustment. In contrast, no significant difference was observed in the risk of noncardiac death among the 3 phenotypes (Table 2). Compared with the poor cardiac condition group, the multimorbidity group showed significantly lower risks of the composite end point, HF hospitalization, all‐cause death, and cardiac death, whereas the risk of noncardiac death did not differ significantly between the 2 groups (Table S6). Kaplan–Meier curves for 2‐year composite end point, hospitalization for HF, all‐cause death, cardiac death, and noncardiac death according to the 3 phenotypes. (A) Composite end point, (B) hospitalization for HF, (C) all‐cause death, (D) cardiac death, and (E) noncardiac death. HF indicates heart failure. Deaths of unknown cause were included in all‐cause death but not included in cardiac or noncardiac death. Table 2Risk of End Points According to the PhenotypeEnd pointPhenotype 1: robustPhenotype 2: poor cardiac condition P valuePhenotype 3: multimorbidity P valuenHRnHRnHR(%)(95% CI)(%)(95% CI)(%)(95% CI)Composite end point194 (24.7)Reference210 (42.9)1.96 (1.58–2.42)<0.001538 (43.3)1.54 (1.28–1.84)<0.001HF hospitalization146 (18.6)Reference156 (31.8)2.00 (1.57–2.54)<0.001394 (31.7)1.54 (1.25–1.90)<0.001All‐cause death75 (9.5)Reference95 (19.4)2.00 (1.42–2.81)<0.001253 (20.4)1.62 (1.21–2.17)<0.001Cardiac death37 (4.7)Reference58 (11.8)2.21 (1.41–3.47)0.001128 (10.3)1.57 (1.05–2.34)0.028Noncardiac death29 (3.7)Reference27 (5.5)1.53 (0.84–2.79)0.16592 (7.4)1.51 (0.93–2.45)0.099HRs were adjusted for age, sex, pathogenesis of HF (ischemic cardiomyopathy or not), atrial fibrillation, as well as body mass index, serum sodium level, and hemoglobin levels at discharge. Deaths of unknown cause were included in all‐cause death but not included in cardiac or noncardiac death.HF indicates heart failure; and HR, hazard ratio. Risk of End Points According to the Phenotype HRs were adjusted for age, sex, pathogenesis of HF (ischemic cardiomyopathy or not), atrial fibrillation, as well as body mass index, serum sodium level, and hemoglobin levels at discharge. Deaths of unknown cause were included in all‐cause death but not included in cardiac or noncardiac death. HF indicates heart failure; and HR, hazard ratio. Internal Validation In the internal validation using the simple split approach, some differences in the distribution of individual I NEED HELP components were observed, particularly for the small‐sized phenotype (phenotype 2), likely reflecting instability in class assignment in smaller subgroups. Nevertheless, the training and test sets showed baseline characteristics (Tables S7 and S8) and distributions of the I NEED HELP components across phenotypes (Figure S4A–S4C) that were broadly consistent with those observed in the entire data set. Kaplan–Meier survival curves in both the training and test sets (Figure S5A–S5F) demonstrated worse prognosis in the poor cardiac condition and multimorbidity groups compared with the robust group, consistent with the entire data set. Notably, in the bootstrap internal validation, the Bayesian information criterion decreased substantially from 2 to 3 classes, with more modest improvements thereafter, supporting the 3‐class solution (Table S9). The agreement in class assignments between the bootstrap replications and the original model, assessed using the Jaccard index, was 0.471 (95% CI, 0.297–0.644; Figure S6). DISCUSSION While the I NEED HELP mnemonic is recognized for its clinical relevance in identifying advanced HF, the interrelationships among its individual components and their practical applicability in real‐world settings have remained insufficiently understood. To the best of our knowledge, this is the first study to provide data on phenotyping of patients with advanced HF based on the criteria. The novel approach of this study, including the use of LCA, revealed major findings. First, ≈85% of the patients hospitalized with acute HF in this registry met at least 1 of the criteria. Second, some of I NEED HELP criteria were slightly related to each other, especially between ‘LVEF < 20% after the HF signs and symptoms stabilized’ and ‘previous or ongoing need of inotropes’ or ‘inability to up‐titrate or need to reduce/discontinue prognostic medications’ and ‘edema or escalating diuretics'. Third, the LCA successfully classified patients into 3 phenotypes, each representing distinct clinical characteristics and outcomes. Patients with phenotypes 2 and 3 had a higher risk of clinical adverse events than those with phenotype 1. Finally, the patterns of GDMT prescription varied substantially across phenotypes, even in GDMT‐eligible patients. Previous studies, including the HELP‐HF (Assessment of the I Need Help Markers in Heart Failure) registry study, which included hospitalized and outpatient populations, reported that ≈20% of patients met at least 1 I NEED HELP criterion. 2 , 10 In contrast, we found that the majority of hospitalized patients with decompensated HF fulfilled at least 1 I NEED HELP criterion, suggesting the need of further risk stratification during this vulnerable phase of HF. Network analysis applied to the criteria demonstrated interconnectivity among the variables. Specifically, a certain association was identified between ‘inability to up‐titrate or need to reduce/discontinue prognostic medications’ and ‘edema or escalating diuretics', suggesting a clinical phenotype characterized by treatment refractoriness and worsening congestion, in association with GDMT intolerability. These findings highlight the imperative need for more sophisticated classification strategies in advanced HF, extending beyond criterion counts, 2 , 10 to adequately address the inherent heterogeneity within this population. Here, we identified 3 mutually exclusive phenotypes from our data set; thus, our findings address this knowledge gap by providing a clinical framework to better characterize this unique patient population. Phenotype 1 (robust) had the lowest number of applicable criteria with better renal function. Given that a prior study found that a lower number of I NEED HELP criteria was associated with a better prognosis, 10 this phenotype may represent an early stage of advanced HF. Conceptually, the I NEED HELP criteria aim to change referral patterns for advanced HF toward “too early” rather than “too late.” 9 Accordingly, patients within this phenotype are not likely to be identified as having truly advanced HF under existing definitions. The patients with this phenotype likely require multidisciplinary interventions to prevent progression to an advanced stage. In particular, the current guideline and statement emphasize that the optimal implementation of GDMT, especially in HF with reduced ejection fraction, is the cornerstone of treatment for advanced HF. 7 , 8 Nonetheless, we found suboptimal implementation of GDMT, especially with renin–angiotensin system inhibitors and MRAs in this phenotype. A previous study demonstrated that the suboptimal implementation of GDMT is driven by a complex interplay of contributors, including an increased risk of adverse drug effects, patient preferences and beliefs, and clinical inertia. 25 These highlight the ongoing need for effective strategies to enhance the widespread use and optimal dosing of GDMT in clinical practice. Patients with phenotype 2 (poor cardiac condition) were younger, had severe cardiac dysfunction, and exhibited higher applicable markers, whereas the number of comorbidities was lower than that of the other phenotypes. This subpopulation had the highest number of I NEED HELP criteria, and therefore, a worse clinical prognosis. Current clinical guidelines recommend considering advanced HF therapies (eg, left ventricular assist device or heart transplantation) when HF stages progress despite optimal implementation of both GDMT and device therapies, leading to uncontrolled clinical symptoms or end‐organ dysfunction. 7 , 8 Given the potential for catastrophic outcomes due to delayed treatment interventions in patients with this phenotype, attending physicians should, under optimized GDMT, promptly consider the eligibility for these advanced HF therapies unless contraindicated and ensure timely referral to an advanced HF center. In addition, the prevalence of dilated cardiomyopathy was relatively higher in this phenotype, and many younger patients may be genetic carriers, which may contribute to their clinical trajectory. More detailed characterization of HF pathogenesis and genetic information may influence the observed phenotypic differences beyond the I NEED HELP criteria, warranting further investigation. Patients with phenotype 3 (multimorbidity) were the oldest, had the highest number of comorbidities, and had better preserved systolic function. Due to its poor prognosis, which is comparable with that of phenotype 2, this phenotype constitutes a significant subset of high‐risk patients with advanced HF based on I NEED HELP criteria. The poor prognosis of the phenotype could be partly explained by noncardiac comorbidities and advanced age, which have a high risk of clinical adverse events driven by the deterioration of both cardiac and noncardiac conditions. 26 , 27 In addition, several major concerns exist about this phenotype. First, compared with the other 2 phenotypes, they had lower eligibility for GDMT and possessed more clinical features (eg, advanced age or severe renal dysfunction) that made them unsuitable for advanced HF therapies, 1 resulting in a limited scope for HF‐specific interventions. Second, given the increased prevalence of HF and multimorbidity over the past 20 years, 28 mainly in developed countries with aging populations, the prevalence of this phenotype is expected to rise in the future. Thus, there is an urgent need to establish a comprehensive and multidimensional framework for managing both cardiac and noncardiac conditions. Although an integrated network of regional HF units and specialized HF centers could be proposed to seamlessly deliver care to these patients, its implementation remains in the early stages of development. 29 To address these challenges, further research is crucial for developing effective interventions. Study Limitations Several limitations must be acknowledged for a thorough interpretation of these results. First, given the potential cohort specific bias, external validation is needed to determine whether the identified phenotypes are reproducible beyond this registry. Although the classification system showed moderate reproducibility in internal validation using bootstrap resampling (Jaccard index, 0.471) and simple split method, external validation was not conducted and remains an important future step. In particular, given the inherent characteristics of LCA, the number of phenotypes may vary according to each analyzed cohort, and different patterns of phenotypes may be observed when analyzing different cohorts. Second, several components of the I NEED HELP criteria were operationalized using pragmatic substitutions based on available registry data, which may have introduced some degree of misclassification. For example, we did not collect the data on >1 hospitalization with HF in the past 12 months, and instead considered any previous HF hospitalization, which may have led to an overestimation. In addition, key components such as natriuretic peptide levels could not be reliably assessed due to substantial missing data. However, such pragmatic adaptations may enhance the applicability of the I NEED HELP framework in real‐world clinical settings where complete data are often unavailable. Future studies are warranted to validate these definitions against more granular data and to refine the classification accordingly. Third, this study was performed on the basis on the Japanese HF registry with unique patient characteristics (eg, older age and low body mass index) and management (eg, frequent use of tolvaptan); therefore, our findings may not be applicable to other countries. Nevertheless, these patterns are real and clinically meaningful within the Japanese setting. The loop diuretic doses in our cohort were consistent with those in other trials or registries in Japan 30 , 31 but lower than those observed in Western countries. 2 , 18 This discrepancy most likely depends on differences in patients’ physical size. Notably, Japanese patients with HF had a poor prognosis, even among those treated with ≥40 mg/d of oral furosemide at discharge. 31 Furthermore, the proportion of patients with edema or escalating diuretics was 62%, which is nearly identical to the 57% observed in the HELP‐HF registry. 10 Fourth, our study had an observational cohort design, and despite adjustment using various prognostic factors, unmeasured or unknown variables, as well as variables not included due to substantial missingness, may have influenced the outcomes. Finally, the study did not develop a parsimonious bedside score to directly assign individuals to a phenotype. Creating such a score would require a separate prediction‐model framework with derivation, validation, and calibration. Instead, we provide class‐defining probabilities and outcome rates (Figure S7), enabling clinicians and researchers to approximate class membership using the 9 I NEED HELP variables and to understand the associated prognostic risk. CONCLUSIONS We identified 3 novel phenotypes with significantly different prognosis among patients with advanced HF on the basis of the I NEED HELP criteria. This classification facilitates risk stratification and supports therapeutic decision making for patients with advanced HF. Sources of Funding The West Tokyo Heart Failure Registry was supported by a grant from the Japan Agency for Medical Research and Development (S.K., 201439013C), Grants‐in‐Aid for Scientific Research (T.Y., JSPS KAKENHI, 23591062, 26 461 088; T.K., 17K09526, 20K08408, 23K07584; A.G., 21K08087), a Grant‐in‐Aid for Young Scientists (Y.S., JSPS KAKENHI, 18K15860), a Grant‐in‐Aid for Clinical Research from the Japanese Circulation Society (Y.S., 2019), a Grant‐in‐Aid from the Japanese Ministry of Health, Labour and Welfare (S.K., H29‐Refractory Disease‐034), a Health Labour Science Research Grant (S.K., 14 528 506), and Sakakibara Clinical Research Grant for the Promotion of Science (T.Y., 2012–2020). Disclosures Dr Kohno received honoraria from Otsuka Pharmaceutical Co. Ltd. and Bayer Pharmaceutical Co. Ltd. and Nippon Boehringer Ingelheim Co. Ltd. Dr Shiraishi received research grants from the SECOM Science and Technology Foundation and the Uehara Memorial Foundation, as well as honoraria from Otsuka Pharmaceutical Co. Ltd. and Ono Pharmaceutical Co. Ltd. Dr Kohsaka received an unrestricted research grant from the Department of Cardiology at Keio University School of Medicine, Bayer Pharmaceutical Co. Ltd., Daiichi Sankyo Co. Ltd., Novartis Pharmaceutical Co. Ltd., and AstraZeneca Pharmaceutical Co. Ltd. The other authors declare that they have no conflicts of interest. There are no patents, products in development, or marketed products to declare. Supporting information Tables S1–S9 Figures S1–S7 STROBE Statement Acknowledgments The authors are grateful to the investigators of the WET‐HF. This manuscript was sent to Amgad Mentias, MD, Associate Editor, for review by expert referees, editorial decision, and final disposition. Supplemental Material is available at https://www.ahajournals.org/doi/suppl/10.1161/JAHA.126.049046 For Sources of Funding and Disclosures, see page 10. References Associated Data Supplementary Materials Tables S1–S9 Figures S1–S7 STROBE Statement