Predicting Survival of End Stage Heart Failure Patients Receiving HeartMate-3 LVAD with Machine Learning. An STS-INTERMACS Analysis.

In the United States the Heartmate3 (HM3) has become the only durable LVAD for implantation in end stage heart failure (HF) patients. Prior predictive scores included older generation devices and thus may not applicable. The purpose of this study was to develop a one-year survival predictive score for HM3 candidates. Adult patients registered in the STS-INTERMACS database, who received primary implant with a HM3 device between 1/1/2014 and 6/1/2020 were included in this study. Epidemiological, clinical, hemodynamic, and echocardiographic characteristics were analyzed. Machine learning, elastic-net method, was used to build a one-year mortality predictive model on the derivation cohort; and its performance was tested on the validation cohort. A total of 3642 patients met inclusion criteria. Of these, 429 died within 1 year after HM3 implantation. Non-survivors were older (61.7 vs. 56.8 years, p<0.001), less frequently male (15.2% vs. 20.6%, p=0.01), had higher frequency of chronic HF (82.8% vs. 79% p=0.004), had a higher frequency of ECMO (10.3 %vs. 4.26%, p<0.001), higher rates of dialysis (6.29% vs. 1.93%, p<0.001), higher frequency of previous cardiac surgery (35% vs. 24.2%, p<0.001), and higher frequency of concomitant cardiac surgery (50.4% vs. 44.3%, p =0.02). Machine learning identified 9 variables associated with one-year mortality in the training data set (Age, right atrial pressure, MELD-XI score, INTERMACS profile 1, Chronic HF, White blood cell count, hemoglobin, heart rate < 90, and ECMO prior to implant), with an AUC 0.68 (figure 1A). The importance of each variable in the predicting model is shown in figure 1B. Our score predicted one year survival in the testing cohort with a good accuracy (AUC:0.72, Figure 1C). Using the machine learning algorithm, we developed a model for predicting one-year survival for patients receiving HM3 LVAD. Age, right atrial pressure and MELD-XI score were the most important predictive factors. [ABSTRACT FROM AUTHOR]