Mortality at 1 yr

To examine the difference in cumulative mortality between the intervention groups, Kaplan–Meier analysis was used to estimate the non-parametric 1 yr survival curves for both intervention groups. The difference between the two survival curves was tested using the log-rank test. The relative risk of 1 yr mortality comparing the heavier sedation with the lighter sedation intervention group was evaluated through estimated hazard ratios from the semi-parametric Cox proportional hazards model, with corresponding confidence intervals. The main model used intervention group assignment as the primary predictor and included the stratification variables of age and MMSE at baseline as covariates. Additional models were constructed to explore for the potential impact of interaction between intervention and baseline Charlson co-morbidity index (CCI), and baseline and in-hospital POD-related (e.g. incident cases, number of days with POD, mean DRS-R-98 severity score) variables as potential independent predictors, on the mortality outcome during the 1 yr follow-up. The proportional hazards assumption was evaluated by examining the intervention group by survival time interaction term in the proportional hazards model and by examining the Schoenfeld residuals.