Machine learning-based models have a greater discriminatory ability compared with conventional regression-based models in predicting likelihood of myocardial recovery in patients with left ventricular assist device (LVAD) support, according to study results published in Circulation: Heart Failure.

Researchers used data from the Interagency Registry for Mechanically Assisted Circulatory Support (INTERMACS) to enroll patients who were aged 18 years and older with heart failure and who received durable continuous-flow mechanical circulatory support from 2008 through 2017. A total of 20,270 patients were included in the study. The primary outcome was LVAD explant for myocardial recovery indication.

Researchers extracted 98 raw clinical variables from the INTERMACS dataset to include for feature selection. The machine learning models were developed in the training cohort (70%) and were assessed in the validation cohort (30%).


Continue Reading

Of the 98 variables, 28 with nonzero coefficients were selected for machine learning model training. Of these, 14 features had a positive association with LVAD-induced myocardial recovery, including bridge-to-recovery implant strategy, current tobacco use, postpartum cardiomyopathy, and recent cardiac diagnosis (1 month to 1 year). In addition, 14 features had a negative association with LVAD-induced myocardial recovery, including use of an implantable cardioverter defibrillator, postimplant left ventricular ejection fraction (0%-20%), and right ventricular assist device implantation with LVAD.

The researchers developed 5 machine learning models, including Bayesian logistic regression (B-LR), linear support vector machine, gradient boosted decision tree, neural network, and random forest. All these models showed the ability to predict LVAD-induced myocardial recovery in the validation cohort with area under the curve (AUC) of greater than 0.810.

The discriminatory ability of the machine learning models was significantly better than that of the regression-based INTERMACS recovery scores, including INTERMACS Cardiac Recovery Score (I-CARS) and INTERMACS Recovery Score (I-TOPS), which had AUCs of less than 0.750 (all P <.001).

The researchers also conducted an additional multivariable logistic regression analysis in the training data set and developed a new INTERMACS LVAD recovery risk score as the I-CARS and I-TOPS scores had been derived from earlier versions of INTERMACS data. The discriminatory ability of the new INTERMACS LVAD recovery risk score (AUC, 0.796) was superior to that of I-CARS and I-TOPS, but it was inferior to the top-performing B-LR machine learning model (AUC, 0.824) in the validation data set (P =.046).

The cumulative incidence of LVAD explantation for myocardial recovery was significantly increased among patients who were predicted to recover in machine learning models vs those who were not (5.1%, 11.5%, 15.8%, and 18.8% vs 0.2%, 1.4%, 1.9%, and 2.6% at 1, 2, 3, and 4 years of LVAD support, respectively; log-rank P <.001).

As study limitations, the researchers noted that their analysis was limited to clinical variables from the INTERMACS registry, and that echocardiographic or radiographic image data were not available. Also, the machine learning models were validated internally but not externally owing to the small number of recovery patients at any given center. In addition, newer devices such as Heartmate 3 LVAD were not well-represented in the cohort.

“Machine learning tools can help the care team to better identify patients who are likely to recover on LVAD support so that the recovery efforts could be maximized on these individuals,” the researchers noted.

Disclosure: Some of the study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.

Reference

Topkara VK, Elias P, Jain R, Sayer G, Burkhoff D, Uriel N. Machine learning-based prediction of myocardial recovery in patients with left ventricular assist device support. Circ Heart Fail. Published online December 24, 2021. doi:10.1161/CIRCHEARTFAILURE.121.008711