The chronic heart failure evolutions: Different fates and routes

Abstract

Aims: Individual prognostic assessment and disease evolution pathways are undefined in chronic heart failure (HF). The application of unsupervised learning methodologies could help to identify patient phenotypes and the progression in each phenotype as well as to assess adverse event risk.

Methods and results: From a bulk of 7948 HF patients included in the MECKI registry, we selected patients with a minimum 2-year follow-up. We implemented a topological data analysis (TDA), based on 43 variables derived from clinical, biochemical, cardiac ultrasound, and exercise evaluations, to identify several patients' clusters. Thereafter, we used the trajectory analysis to describe the evolution of HF states, which is able to identify bifurcation points, characterized by different follow-up paths, as well as specific end-stages conditions of the disease. Finally, we conducted a 5-year survival analysis (composite of cardiovascular death, left ventricular assist device, or urgent heart transplant). Findings were validated on internal (n = 527) and external (n = 777) populations. We analyzed 4876 patients (age = 63 [53-71], male gender n = 3973 (81.5%), NYHA class I-II n = 3576 (73.3%), III-IV n = 1300 (26.7%), LVEF = 33 [25.5-39.9], atrial fibrillation n = 791 (16.2%), peak VO₂% pred = 54.8 [43.8-67.2]), with a minimum 2-year follow-up. Nineteen patient clusters were identified by TDA. Trajectory analysis revealed a path characterized by 3 bifurcation and 4 end-stage points. Clusters survival rate varied from 44% to 100% at 2 years and from 20% to 100% at 5 years, respectively. The event frequency at 5-year follow-up for each study cohort cluster was successfully compared with those in the validation cohorts (R = 0.94 and R = 0.84, P < 0.001, for internal and external cohort, respectively). Finally, we conducted a 5-year survival analysis (composite of cardiovascular death, left ventricular assist device, or urgent heart transplant observed in 22% of cases).

Conclusions: Each HF phenotype has a specific disease progression and prognosis. These findings allow to individualize HF patient evolutions and to tailor assessment.

Keywords: Cardiopulmonary exercise test; Heart failure; Prognosis; Topological data analysis.

Figure 1

Study workflow. The block diagram describes the implemented analysis which are categorized in three main steps: data pre‐processing, the unsupervised analysis (cluster identification, cluster characterization and trajectories inference), and the supervised analysis (survival analysis and validations).

Figure 2

Cluster identification by topological data analysis. Each node of the network represents a group of patients. The node size is proportional to the number of enclosed samples, while edges thickness depicts the Jaccard's index (i.e., the number of samples shared by nodes). Each identified cluster has been highlighted with different colors.

Figure 3

Clusters characterization. (A) The bubble plot represents the average frequency of samples exhibiting a specific categorical feature (y‐axis) in each cluster (x‐axis); the higher the circle, the higher the sample rate. (B) The bubble plot represents the distribution of continuous variables (min‐max scaling); the higher the circle size, the higher the average feature value in a specific cluster. All frequencies, means and standard deviations of each feature for each cluster are present in Table S1.

Figure 4

Trajectory Analysis. (A) The dot plot shows the UMAP representation of TDA network nodes which allowed to inferred cluster centroids (red squares for bifurcation points, yellow diamonds for stopping point and black circles for all other clusters) and to identify the most likely trajectories between clusters. The arrows depict the HF evolution albeit a reverse evolution is possible as shown in Figure S3A, B. (B) The boxplot represents the distribution of pseudo‐time estimated for each cluster; dots represent the pseudo‐time value calculated for each TDA node.

Figure 5

Survival analysis. (A) The Kaplan–Meier curve of the study population with the event‐free survival for each cluster at 5‐year follow‐up. (B) The Kaplan–Meier curve of the two validation cohorts combined.

Figure 6

Validation analysis. (A) The scatter plot shows the correlation between the frequency of study endpoints at 5 years within each cluster in the study population (x‐axis) and in the internal validation cohort (y‐axis). (B) The same plot is shown to correlate frequency of study endpoints at 5 years, even with the external validation cohort (y‐axis). For both the correlation analysis, the Pearson's correlation coefficient (R) and the relate P‐value were computed to evaluate the strength of association between the two methodologies. The 95% confidence interval of the trendline (orange line) is depicted in light orange.