Author:
Martin Glen P.,Hindricks Gerhard,Akbarov Artur,Kapacee Zoher,Parkes Le Mai,Motamedi-Ghahfarokhi Golnoosh,Ng Stephanie,Sprague Daniel,Taleb Youssef,Ong Marcus,Longato Enrico,Miller Christopher A.,Shamloo Alireza Sepehri,Albert Christine,Barthel Petra,Boveda Serge,Braunschweig Frieder,Johansen Jens Brock,Cook Nancy,de Chillou Christian,Elders Petra J.M.,Faxen Jonas,Friede Tim,Fusini Laura,Gale Chris P.,Jarkovsky Jiri,Jouven Xavier,Junttila Juhani,Kiviniemi Antti,Kutyifa Valentina,Lee Daniel,Leigh Jill,Lenarczyk Radosław,Leyva Francisco,Maeng Michael,Manca Andrea,Marijon Eloi,Marschall Ursula,Vinayagamoorthy Manickavasagar,Nielsen Jens Cosedis,Olsen Thomas,Pester Julie,Pontone Gianluca,Schmidt Georg,Schwartz Peter J.,Sticherling Christian,Suleiman Mahmoud,Taborsky Milos,Tan Hanno L.,Tflt-Hansen Jacob,Tijssen Jan G.P.,Tomaselli Gordon,Verstraelen Tom,Warnakula Olesen Kevin Kris,Wilde Arthur A.M.,Willems Rik,Willems Dick L.,Wu Katherine,Zabel Markus,Peek Niels,Dagres Nikolaos
Abstract
AbstractIntroductionSudden cardiac death (SCD) is the leading cause of death in patients with myocardial infarction (MI) and can be prevented by the implantable cardioverter defibrillator (ICD). Currently, risk stratification for SCD and decision on ICD implantation are based solely on impaired left ventricular ejection fraction (LVEF). However, this strategy leads to over- and under-treatment of patients because LVEF alone is insufficient for accurate assessment of prognosis. Thus, there is a need for better risk stratification. This is the study protocol for developing and validating a prediction model for risk of SCD in patients with prior MI.Methods and AnalysisThe EU funded PROFID project will analyse 23 datasets from Europe, Israel and the US (∼225,000 observations). The datasets include patients with prior MI or ischemic cardiomyopathy with reduced LVEF<50%, with and without a primary prevention ICD. Our primary outcome is SCD in patients without an ICD, or appropriate ICD therapy in patients carrying an ICD as a SCD surrogate. For analysis, we will stack 18 of the datasets into a single database (datastack), with the remaining analysed remotely for data governance reasons (remote data). We will apply 5 analytical approaches to develop the risk prediction model in the datastack and the remote datasets, all under a competing risk framework: 1) Weibull model, 2) flexible parametric survival model, 3) random forest, 4) likelihood boosting machine, and 5) neural network. These dataset-specific models will be combined into a single model (one per analysis method) using model aggregation methods, which will be externally validated using systematic leave-one-dataset-out cross-validation. Predictive performance will be pooled using random effects meta-analysis to select the model with best performance.Ethics and disseminationLocal ethical approval was obtained. The final model will be disseminated through scientific publications and a web-calculator. Statistical code will be published through open-source repositories.
Publisher
Cold Spring Harbor Laboratory