Association between type of bystander cardiopulmonary resuscitation and survival in out-of-hospital cardiac arrest: A machine learning study.

Aim: In the event of an out of hospital cardiac arrest (OHCA) it is recommended for a sole untrained bystander to perform compression only CPR (CO-CPR). However, it remains unknown if CO-CPR is inferior to standard CPR (S-CPR), including both compressions and ventilation, in terms of survival. One could speculate that due to the current pandemic, bystanders may be more hesitant performing mouth-to-mouth ventilation. The aim of this study is to assess the association between type of bystander CPR and survival in OHCA. Methods: This study included all patients with a bystander treated OHCA between year 2015-2019 in ages 18-100 using The Swedish Registry for Cardiopulmonary Resuscitation (SRCR). We compared CO-CPR to S-CPR in terms of 30-day survival using a propensity score approach based on machine learning adjusting for a large number of covariates. Results: A total of 13,481 patients were included (5,293 with S-CPR and 8,188 with CO-CPR). The matched subgroup consisted of 2994 cases in each group.Gradient boosting were the best models with regards to predictive accuracy (for type of bystander CPR) and covariate balance. The difference between S-CPR and CO-CPR in all 30 models computed on covariate adjustment and 1-to-1 matching were non-significant. In the 30 weighted models, three comparisons (S-CPR vs. CO-CPR) were significant in terms of improved survival; odds ratio for men was 1.21 (99% confidence interval (CI) 1.02-1.43; Average treatment effect (ATE)); for patients ≥73 years 1.57 (99% CI 1.17-2.12) for Average treatment effect on treated (ATT) and 1.63 (99% CI 1.18-2.25) for ATE. Remaining 27 models showed no differences. No significances remain after adjustment for multiple testing. Conclusion: We found no significant differences between S-CPR and CO-CPR in terms of survival, supporting current recommendations for untrained bystanders regarding CO-CPR.