Right Ventricular Failure Accurately Predicted By Bayesian Statistical Model

A Bayesian prognostic model of right ventricular failure (RVF) provided highly accurate predictions of acute, early, and late RVF, based on a large, multicenter INTERMACS registry, according to results published in JACC: Heart Failure.

Researchers investigated the use of a Bayesian statistical model to address the predictive capacity of existing risk scores, which are currently rather limited. They considered inter-relationships and conditional probabilities among independent variables to achieve sufficient statistical accuracy.

They noted that the Bayesian analysis “provides the possibility to examine the changes in risk over time, as well as explore hypothetical ‘what if’ scenarios by entering variables manually. One could even envision entering variables inter-operatively, thereby acknowledging recent reports of the importance of intra-operative events to the occurrence of post-operative RV failure.”

The study included data from 10 909 adult patients from the INTERMACS registry (Interagency Registry for Mechanically Assisted Circulatory Support) who had primary left ventricular assist device (LVAD) implantation from December 2006 to March 2014.

An initial set of 176 pre-implant variables was complied as well as a separate, tree-augmented Naïve Bayes model for each post-implant RVF end point. These end points were categorized as acute (<48 hours), early (48 hours to 14 days), and late (>14 days).

The acute RVF model developed consisted of 33 variables, including systolic pulmonary artery pressure (PAP), white blood cell count, left ventricular ejection fraction, cardiac index, sodium levels, and lymphocyte percentage.

The early RVF model had 34 variables, including systolic PAP, pre-albumin, lactate dehydrogenase, INTERMACS profile, right ventricular ejection fraction, pro-B-type natriuretic peptide, age, heart rate, tricuspid regurgitation, and BMI. The late RVF model had 33 variables, mostly predicted by peripheral vascular resistance, model for end-stage liver disease (MELD) score, albumin, lymphocyte percentage, mean PAP, and diastolic PAP.

The Bayesian models had accuracy scores between 91% and 97%, areas under the curve between 0.83 and 0.90, sensitivities of 90%, and specificity between 98% and 99%, all of which significantly outperformed current risk scores.

“The Bayesian models reported here are particularly suited for combining large sets of risk factors because they are based on conditional probabilities of the likelihood of RVF for a given combination of inter-related variables,” the authors noted. “In this way, these algorithms better reflect human logic in prioritizing dynamic clinical information, yet benefitting from the corpus evidence provided by the INTERMACS registry.”

They also noted that in a clinical setting, the model can predict the likelihood of RVF even with a limited or incomplete set of data. Additionally, if more data points are added, the algorithm’s predictive ability improves incrementally.

“We recognize that the utility of these Bayesian models, containing over 30 variables, will depend greatly on the ease/difficulty by which it can be calculated,” the authors concluded. “For this reason, our ongoing work aims to provide an accessible and easy-to-use decision support tool for physicians and patients engaged in LVAD discussion.”

Reference

Loghmanpour NA, Kormos RL, Kanwar MK, et al. A Bayesian model to predict right ventricular failure following left ventricular assist device therapy. JACC Heart Fail. 2016. doi:10.1016/j.jchf.2016.04.004.