Maayke Hunfeld

178 Chapter 6 differences were analyzed with Chi-square test or Fisher’s exact test when applicable. Continuous data was presented as median and interquartile ranges (IQR) for skewed data, and mean and standard deviation (SD) for normal distributed data. Differences were tested using an independent sample t-test for continuous data or Mann– Whitney U test dependent on normality. The associations of first documented rhythm, AED use, bystander BLS, year of event and the post AED guideline change period with long-term neurologic outcome were calculated with a multivariable logistic regression model. The choice of inclusion of covariates was made in three steps. First, the following covariates were considered based on existing literature: age, gender, pre-existing condition (yes or no and related to CPR event or not), SES (1, 2 or 3), event location (private or public), year of event (including before and after the AED guideline change), witnessed arrest (yes or no), bystander CPR (yes or no), bystander AED use (yes or no), CPR duration (in minutes), first documented cardiac arrest rhythm (shockable, non-shockable or unknown), cause of arrest (specific), ECPR (yes or no) and first lactate and pH after ROSC. Second, collinearity analysis to explore correlation between all covariates using a correlation matrix was performed. A cut-off value of >0.7 was used for the exclusion of variables in the model. Third, inclusion of the abovementioned potential confounders in the final models was based on >10% change of the effect estimate in the crude model. These covariates were entered one-by-one in the crude model to see the effect on the effect estimate. Results are presented as odds ratio (OR) and 95%-confidence interval (CI). A sensitivity analysis comparing the different definitions of favorable neurologic outcome (PCPC 1-2 vs PCPC 1-3, or no pre-and post-arrest difference) was performed. Stratified analysis by age group (infant; aged <1 year, child; aged 1 to 11 years and adolescent; aged 12 to 18 years as well as below and above 8 years of age) was also done. Lastly, a propensity score analysis using 1:1 nearest-neighbor matching of shockable to non-shockable rhythm was performed. The propensity score was estimated using a multivariable logistic regression model including the following variables: gender, age at arrest and year of event. Both groups were tested for association with long-term neurologic outcome using a multivariable logistic regression model. Our data contained missing values for CPR duration (19%). Other covariates had < 10% missing data. Variables were imputed using multiple imputation (n = 5 imputations) function based on the distribution of existing data. A two-tailed p-value < 0.05 was considered statistically significant. All analyses were conducted using SPSS software version 24 (IBM SPSS Statistics for Windows, Armonk, New York, USA).