Matt Harmon

161 Chapter eight Logistic regressions are presented as odds ratios (OR) with 95% confidence intervals (CI) with OR below 1 indicating better, and above 1 indicating worse neurological outcome. OR relating to continuous data describe the variation per one unit (1 kPa for PaCO 2 and 1 unit for pH). For s-Tau analysis, multivariable linear regression was used including the same co-variables and interaction analyses as above. S-Tau values where transformed to a logarithmic scale and used as outcome in the linear regression analyses. The regression coefficients achieved for each independent variable were transformed back and reflect the multiplicative change in s-Tau. This means that coefficients below 1 correspond to a decrease in s-Tau and above 1 to an increase. Linear regressions are presented as coefficient estimates with 95% CI. Multiple imputation was used to compensate for missing values; 20 imputations were generated using chained equations and evaluated by graphical methods. The estimates from the logistic and linear regression for each imputed sample were combined into one estimate with 95% CI including the uncertainty from the multiple imputations. The primary analyses were performed on a multiple imputation cohort. A complete case cohort was used for sensitivity analysis. We regarded a two-sided P-value < 0.05 as significant. Analyses were conducted using IBM SPSS statistics for Windows version 22.0, Armok NY and R:A language and environment for statistical Computing version 3.3.3R Foundation for Statistical Computing, Vienna, Austria and the package mice was used for multiple imputations. 30 Results From the 939 patients included in the TTM-trial we excluded patients who did not survive the analysis period (n = 62), patients with no PaCO 2 data (n = 2) and patients with no data on neurological outcome (n = 6) (Figure 1), leaving 869 (93%) patients for analysis. Additional data on the number of excluded patients at each measuring point are displayed in Supplementary Table 1, Additional File 1. Baseline characteristics of the included patients are presented in Table 1. 878 of 6952 analyzed measuring points were missing (12.6%), detailed information regarding the number of missing values at each time point is shown in Table 2 and multiple imputation was used to overcome the missingness. 485 patients had no missing values. Overall, 440 (50.6%) of 869 patients had a good outcome while the outcome of 429 (49.4%) was considered poor.