Categorical variables are expressed as proportions and compared Crizotinib with the Mantel-Haenszel chi-squares test or Fisher’s exact test.For the analysis of RRT dose versus outcome, CRRT and IRRT patients were analysed separately because of well-recognised differences between these populations in observational studies [21,22]. Exploratory univariate analysis for several variables was performed to identify possible risk (or protective) factors associated with ICU mortality. Multivariable logistic regression analysis was then conducted to test the relationship between RRT dose and ICU mortality, adjusted for confounding factors.Based on the results of the univariate analysis, the covariates included in the CRRT model were sex, age (10-year increments), sequential organ failure assessment (SOFA) and serum creatinine at CRRT initiation and CRRT downtime (hours); for IRRT the covariates included were age (10-years increments), sex and RIFLE class at IRRT initiation.

In addition to adjusting for significant covariates, residual confounding and selection effects were addressed using propensity scores. We generated a propensity score using multivariable logistic regression with more-intensive RRT dose as the dependent variable, as previously described [22,23]. Variables included in the propensity score were gender, weight, SOFA score and serum creatinine at RRT initiation. We fitted models for ICU mortality only adjusted for covariates and a combination of covariates plus the propensity score. We assessed for collinearity between variables using tolerance and variance inflation factors; there was no significant collinearity detected.

The model’s goodness of fit was tested with the Hosmer-Lemeshow statistic.As sensitivity analyses, RRT dose was evaluated as both continuous variables and categorical variables. As continuous variable for CRRT, we used the actual value or as increments of 10 ml/kg/hour; for IRRT, we used number of IRRT sessions per week (possible range: 1 to 7). As categorical variables, we created RRT dose categories based on the literature, as well as standard statistical groupings (median, tertiles). Posthoc multivariate analyses were also performed limiting the analysis to specific subgroups of CRRT patients (septic patients, by simplified acute physiology score (SAPS) scores, �� 25 hours of CRRT).

Because of the relatively small sample size of the IRRT, subgroup analysis was not performed.Finally, ICU survival by RRT dose categories was presented graphically using Kaplan-Meier product limit survival plot. Two-tailed p values less than 0.05 were considered significant. Statistical analyses were conducted using STATA 10 (StataCorp LP, College Station, TX, USA).ResultsEnrollment and baseline characteristicsCharacteristics Cilengitide of the participating centres are shown in Table 1 in Additional data file 2.