Read on Twitter
VERY excited
to see our paper published @ESC_Journals #ESCHeartFailure: “Clinical phenogroups are more effective than LVEF categories in stratifying heart failure outcomes”
Thread
to see our paper published @ESC_Journals #ESCHeartFailure: “Clinical phenogroups are more effective than LVEF categories in stratifying heart failure outcomes”Thread
We all rely on LVEF for #heartfailure classification while most of us see the downsides: fuzzy cutoff HFrEF/HFpEF, shared epidemiological/pathophys features, great variation in LVEF measurement across methods & observers. @filpapos @trinemoholdt @Filippatos @kevin_damman
LVEF categories started when major HF trials used low LVEF to enrich for outcome events kn their population. But now, HFpEF is increasingly prevalent & HFrEF prognosis has improved dramatically. @AdriaanVoors @BA_Mulder @MihaiTrofenciuc
Unsupervised machine learning looks for undetected patterns in a dataset without human‐labelled data. These methods have superior classification ability compared to uni- or multivariable ‘human’ classifications, esp. in complex datasets e.g. HF patients.
In this study, we applied unsupervised machine learning to 1693 HF patients from the #PACTHF trial to identify clinical phenogroups. We hypothesized that phenotype‐based classification would be more effective than LVEF categories in predicting outcomes. https://jamanetwork.com/journals/jama/fullarticle/2725688
We found 6 clinical phenogroups, each characterized by a predominant comorbidity: coronary heart disease, valvular heart disease, AF, sleep apnoea, COPD, or few comorbidities.
On the other hand, LVEF showed a linear relation with most variables (
sex, age, blood pressure, presence of AF, and presence of hypertension, 
NT‐proBNP and history of coronary revascularization).
sex, age, blood pressure, presence of AF, and presence of hypertension, NT‐proBNP and history of coronary revascularization).
All phenogroups had patients spanning the whole LVEF range. (Of note, our population had LVEF median of 49 (33–58) so HFpEF was well represented in this ‘all comer’ acute HF population).
Finally, phenogroup-based classification better separated outcomes than LVEF categories. Figures show all-cause 
or 
at 6mo, similar results at 12mo and for cardiovascular or HF at 6 and 12 mo.
or at 6mo, similar results at 12mo and for cardiovascular or HF at 6 and 12 mo.
Interesting to see that the ‘valvular’ and ‘COPD’ phenogroups had the worst prognosis, not surprisingly the phenogroup with few comorbidities had better outcomes. (Regardless of LVEF!)
@KDujardin @ShelleyZieroth @BiykemB @KevinShahMD @HFA_President
@KDujardin @ShelleyZieroth @BiykemB @KevinShahMD @HFA_President
Very proud of this transatlantic collaboration, with expert mentoring by @mmamas1973 @JJheart_doc @yaleHFdoc @denniskomd and most importantly the stellar @hvanspall 
Almost forgot the full paper link for full details!! #openaccess #FOAMed #CardioTwitter https://onlinelibrary.wiley.com/doi/full/10.1002/ehf2.13344
