Analysis of implantable cardiac device diagnostics to tailor management of heart failure

Wednesday, November 7, 2018 - 2:30pm - 3:00pm
Lind 305
Tracy Bergemann (Medtronic)
Implantable cardiac devices collect physiologic parameters and heart rhythm metrics continuously on heart failure patients. Remote monitoring services can combine this information with daily blood pressure, weights and symptoms into a risk prediction algorithm and tailor patient treatment. To determine the clinical benefits of using such a risk prediction algorithm to guide heart failure treatment, appropriate models to assess the potential to offset the negative spiral of recurrent morbidity events and ultimately death are needed.

Traditional approaches to model recurrent events in heart failure include Poisson with overdispersion correction, negative binomial, and Anderson-Gill models. The Poisson and negative binomial models do not account for timing of recurrent events. The Anderson-Gill model assumes that the intensity over the multivariate counting process is constant over time after adjusting for covariates and that those covariates have a proportional effect. These assumptions are frequently questioned in practical applications. Moreover, none of these approaches account for the competing risk of death.

The presentation discusses multistate models to depict recurrent heart failure events and death in prospective studies. This method will be compared with the more traditional models described above. The methods are evaluated in simulations under several null and alternative hypothesis scenarios. The utility of said multistate models for the assessment of risk prediction algorithms that deliver personalized heart failure management will be highlighted.