Purpose of the study:

There are various of factors that affect Return of spontaneous circulation (ROSC) after cardiac arrest (CA). The aim of this study was to evaluate the RACA algorithm for Out of Hospital Cardiac Arrest (OHCA) and propose a new method for other registries to re-evaluate their own data sets. 
Materials and methods:
 
15342 records from a 3-year East Midlands Ambulance Service (EMAS) cardiac arrest registry data base between 2014 to 2017 were retrospectively analysed. Excluded were all patients not in cardiac arrest, traumatic CA, ROSC before crew arrival, DNAR patients and dead on scene. After data cleaning, 6121 eligible patients of all ages remained, for which we built a model for the binary classification with 30 independent categorical features and ROSC at any time as our target variable. 

We split the dataset into a training and validation sets. CatBoost (Category and Boosting) is an ensemble algorithm, which is a combination of several decision trees. The result of the final algorithm is a function of the results from base algorithms (decision trees).
Within EMAS our key features can be grouped into crew-, resuscitation-, patient and scene dependant variables. We further analysed a rarely published and key feature in our data set labelled "role and training level of scene".
From the 30 independent features and ROSC at any time as our target variable we selected the 18 key features. 

Conclusion:  

We are proposing a new method for CA registries to revalue their own data sets and identify feature importance. This allows extended analysis of CA data and helps ambulance services to critically evaluate their EMS systems.

4 - Novel gradient boosting method to identify feature importance for ROSC after Cardiac Arrest (RACA)