Objectives
This project aims to create a prognostic model for the CRT population. The goal is to provide estimates on (a) expected EF improvement, (b) cardiac remodeling as shown through left ventricular end-systolic volume (LVESV) and most importantly on (c) cardiovascular and (d) all-cause mortality.
Scientific rationale
ML and AI models open a personalized approach to medicine. Instead of a practice based on guidelines, we model tools that would assist the physician in clinical decision-making based on the individual’s situation. This is a significant paradigm shift for CRT. There are currently 48 risk scores for prognosis in CRT patients. Some are just HF risk scores. The previous iterations of machine learning risk scores had a low number of patients (e.g. two hundred to nine hundred patients) We define prognosis in the CRT population as (1) improvements in ejection fraction (EF), (2) cardiac remodeling, (3) cardiovascular mortality and (4) all-cause mortality.
Primary outcomes
The primary outcomes are ML models predicting (1) LVEF, (2) LVESV and LVEDV at 6 months and (3) cardiovascular and all-cause mortality. They will be evaluated based on the accuracy of the prediction in comparison with the test set. The appropriate accuracy is above 80%.
Secondary outcomes
Secondary outcomes are ML models predicting the time until next hospitalization following CRT device implantation and successful treatment as defined by a 15% reduction in LVESV and/or LVEF improved by at least 5%.
The other secondary objective is to rank the features that have the most influence on predicting LV improvements as well as cardiovascular and all-cause mortality. Some features pertain to the medical history and some features pertain to exams.