Working towards the prediction and prevention of sudden cardiac death in a major project funded by the European Union
New research papers published by Patchs Health data scientists in medRxiv and The European Heart Journal
Challenge
Sudden cardiac death (SCD) is the leading cause of death in Europe. Accounting for approximately 20% of all deaths (between 350,000 to 700,000 a year), it continues to present a major, urgent, and as yet unsolved public health problem.
Although there is an established link between SCD and patients with previous myocardial infarction (MI), the decision of whether to go ahead with the accepted treatment – which involves the implantation of an implantable cardioverter-defibrillator (ICD) – has historically been based solely upon the degree to which any given patient’s left ventricular ejection fraction (LVEF) has been impaired.
Unfortunately, relying on this strategy can lead to both:
- the over-treatment of patients – a dangerous practice, because there are considerable long-term risks associated with carrying an ICD device – and
- to the under-treatment of patients. This is because, paradoxically, the majority of SCD cases occur in those with moderate, not severe, LVEF.
In short, looking at LVEF alone is insufficient for an accurate assessment of SCD likelihood. A new approach to SCD risk stratification was urgently needed. This is where Patchs Health came in.
Solution
In collaboration with the University of Manchester researchers across the EU, our data scientists worked to develop a radically new SCD-risk prediction model. Their work has recently been published in medRxiv and The European Heart Journal.
The first stage of the project involved the analysis of new, highly-phenotyped data. This was necessary because patient care has improved in recent years, and it was clear that the risk of SCD had decreased and that the current data sets were outdated.
The new data was collected from a wide range of sources, including national registries, institutional research databases, electronic health records and claims databases. With around 200,000 patients included, it represented the largest ever study of people with a history of MI.
Our data scientists analysed the new information by stacking eighteen of the data sets into a single database (datastack). For data governance reasons, the remaining data was analysed remotely (remote data). Five analytical approaches were employed in the process of analysis, such that the new risk prediction model could be created.
The analytical approaches used were: (i) the Weibull model, (ii) the flexible parametric survival model, (iii) random forest, (iv) the likelihood boosting machine, and (v) neural network.
Outcome
Our data scientists completely reassessed the role of routine ICD implantation in patients whose LVEF had been reduced by 35% or less. By performing a meta-analysis, they were able to select the most accurate predictive model for SCD likelihood.
Now, with this new model in hand, the PROFID Project has been able to move successfully onto its second stage: clinical trials.
These trials, known as the PROFID EHRA trials, randomly assess the role ICDs currently play in contemporary optimal medical therapy (OMT). They are currently ongoing.
You can check out the latest developments in PROFID here, and read the full research articles here and here.
