Descriptive data were summarized as means, standard devi­ations, and medians for continuous variables and proportions for categorical variables. Univariate analysis between categorical independent variables and each outcome was performed by means of 2 x 2 contingency tables. Variables having statistically significant correlation with the outcomes (p < 0.05) according to the Fisher exact test were included in the regression analysis. Univariate analysis between continuous independent variables and each outcome was described by correlation coefficients. The selection of variables for inclusion in the initial regression model was based on the following criteria: measurements for a particular variable were available for more than 80% of the patients, there was a statistically significant correlation with the outcomes (2-tailed p < 0.20), and the variable was not highly correlated with other variables. If independent variables were highly correlated (i.e., Pearson correlation coefficient > 0.8), only one of the correlated variables was included in the regression model, on the basis of biological plausibility and smallest p value.

Multiple logistic regression was used to assess the strength of independent associations between the independent variables (milrinone dosing, patient characteristics) and the dependent or outcome variables (development of low cardiac output syndrome, arrhythmia, or thrombocytopenia). A stepwise back­ward method was used as the primary variable-selection technique for multiple logistic regression, beginning with all eligible variables. Lists the variables entered into the multiple logistic regression. Milrinone dosing variables were retained in all of the regression models to assess their associations with outcome variables. The choice of a final logistic regression model containing independent variables was based on the C index (where a larger number, approaching 1, indicates greater accuracy of the chosen model) and the Hosmer-Lemeshow test (where a smallerp value indicates lack of goodness of fit). kamagra tablets

Data were abstracted into a customized database (Access, Microsoft). Data manipulation and statistical analyses were performed with SPSS 13.0 for Windows (SPSS Inc, Somers, New York), SAS 9.1 for Windows (SAS Inc, Carey, North Carolina), and Excel (Microsoft). The study protocol was approved by the Research and Ethics Board at The Hospital for Sick Children.