hvsa
Home
Courses
Course Objectives
Programs
CME Credits
Cases and Abstract Submissions
venue and accomodations
HVSA
Directors and Faculty
Videos
Register

 

New York City Conference

Back to 2014 Annual Meeting Abstracts


How Can Complex Statistical Models Help Cardiologists with VHD Patient Prognostication? The Example of Allograft Valve Function to Predict Patient Outcome
Eleni-Rosalina Andrinopoulou, Dimitris Rizopoulos, Emmanuel Lesaffre, Johanna Takkenberg.
Erasmus Medical Center, Rotterdam, Netherlands.

BACKGROUND: In the prediction of prognosis for new patients suffering from severe heart valve disease, cardiologists consider patient characteristics, echocardiographic measurements (e.g. aortic gradient AG and aortic regurgitation AR), to estimate survival and freedom from reoperation. Intuitively, cardiologists adjust their prognosis over time, with the change in clinical status of the patient at each clinic visit. This study aims to illustrate dynamic predictions using all available information.
METHODS: In our study, 270 patients after allograft aortic valve replacement were followed for 20 years. To analyze patient characteristics, longitudinal echocardiographic and survival data and to use this information to derive survival and freedom from reoperation predictions, we used the joint modeling framework. This approach correlates the AG and AR progression with survival and freedom from reoperation in different ways, e.g.: 1) AG and AR values at specific time points, 2) AG and AR values and their directions at specific time points and 3) the whole AG and AR evolutions up to specific time points. Furthermore, different patient characteristics may play an important role. Thus, we derive predictions for a new patient assuming different models. Finally, since selecting a single model ignores model uncertainty we also obtain weighted predictions.
RESULTS: Figure 1, which presents the predictions using different models for a future patient, shows a variation of results we obtained. From the combined predictions we observed that different models are appropriate for different patients but also for different time points within the same patients. Specifically for the future patient, we observe a contribution of Model4 and Model6 in the predictions.
CONCLUSIONS: Every patient is unique and assuming same prediction models for all patients may not be appropriate. The proposed approach provides the cardiologist a useful tool to assess the impact of AG and AR on patient prognosis. Although the concept needs to be investigated further, it could eventually serve as an early warning system, allowing the necessary time for the cardiologist to plan an intervention.


Back to 2014 Annual Meeting Abstracts

     

Home | Courses | Objectives | Program | CME Credit | Cases & Abstracts | Venu & Accomodations | HVSA | Committee & Faculty | Register | Privacy Policy