Towards Personalized Heart Failure Care

Verdonschot and colleagues phenotyped and clustered a large non-ischemic, non-valvular dilated cardiomyopathy (DCM) cohort (N=795 patients, EF<50%) using an unsupervised machine learning approach.

Of 47 clinical variables (demographics, imaging, lab, comorbid disease, genotyping, etc.), 27 were used to define 4 characteristic phenogroups (PG): PG1 (‘mild systolic dysfunction’, 42%) consisted of patients with mildly impaired LVEF, low NT-proBNP, and little symptoms; PG2 (‘autoimmune’, 10%) consisted of patients with autoimmune disease and the majority were females; PG3 (‘arrhythmias’, 21%) had a male predominance, a high prevalence of atrial fibrillation or nsVTs and a high proportion of mutations (e.g. Titin, or Lamin A/C); PG4 (‘severe systolic dysfunction’, 27%) consisted of patients with severely depressed LVEF and symptoms, and dilated LVs.

In a subset of the cohort (N=91), RNAseq was performed and transcriptomic signatures tracked well along phenogroups with metabolism, inflammation, and fibrosis being major enriched pathways. Phenogroups were also associated with hard endpoints Lastly, a decision tree based on 4 variables (EF, autoimmune disease, atrial fibrillation, creatinine) was built and was accurate in stratifying patients into phenogroups in a large proportion in 2 external validation cohorts. Also in these cohorts, PGs were associated with hard outcomes.

The study is remarkable in several aspects. It shows that in a large HF(m)rEF cohort significant phenotypic and transcriptomic heterogeneity exists and that this impacts outcome. For HF with reduced ejection fraction, we do have well-established therapies such as drugs interfering with neurohumoral signaling (beta-blockers, RAS inhibitors, neprilysin inhibitor)1 and new ones just came around the corner such as SGLT2 inhibitors2,3. However, the variables informing treatment decisions are still rather coarse (EF, NYHA class) and genuine heart failure drugs (not re-purposed from other indications such as diabetes) often have a hard time transitioning to clinical success. When comparing the field to other disciplines such as oncology where personalized approaches (e.g. based on mutations, precise tumor phenotyping) are now standard of care in many cancers, cardiology lags behind. This becomes even more apparent when looking at HF with preserved EF, an arguably yet more heterogeneous disease collective than HF(m)rEF, where all large trials failed so far (at most some signals towards benefit in subgroups)4. A similar approach as in the current study has been applied to HFpEF patients participating in the TOPCAT study before5. Stratifying and selecting patients in order to enrich future clinical trials and care is a key issue and may help to obtain positive results in a more efficient way. The approach laid out by Verdonschot and coworkers is one approach to do so.
Intriguingly, the work also demonstrates that despite in-depth phenotyping was done here, a limited set of easily obtainable parameters may also to the job in a high proportion of subjects.
Lastly, the molecular signatures and pathways enriched in phenotypic subsets (here, RNAseq transcriptome patterns such as inflammation, metabolism, ECM remodeling) may aid in devising new treatment targets.