In-hospital Cardiopulmonary Resuscitation during Asystole: Data Analysis
Data were analyzed to determine whether specific interventions employed during an arrest may be associated with improved 24-h survival rates among patients with an initial rhythm of asystole. Categorical variables were analyzed by cross-tabulation against status (alive vs expired) 24 h after initiation of CPR. For each table, the xa statistic was computed to determine whether the association between 24-h outcome and each of the categorical variables was greater than that expected by chance. An alpha level of p<.01 was decided on prior to analysis in an effort to minimize type 1 errors, which can arise when the ratio of subjects-to-variables is less than optimal. Logistic regression analysis was used to identify the independent contributions of each of the various therapeutic interventions to the prediction of 24-h outcome. A regression model which included the main effects of all intervention variables was compared with several reduced models, in which variables with the smallest regression coefficients were successively eliminated from the regression equation. These comparisons resulted in a reduced set of variables, which, taken together, provided an efficient, adequate fit of the data. Continuous variables, ie, CPR duration and age, were analyzed by f-tests between 24-h survivors and nonsurvivors. website
Survivors did not differ statistically from nonsurvivors in age, gender, primary diagnosis, or duration of CPR efforts. However, patients who received norepinephrine drip (N = 43) were more likely to survive than those who did not (39.5 percent vs 14.1 percent; p< .01) and those who received lidocaine drip (N = 21) were more likely to survive than those who did not (47.6 percent vs 18.2 percent; p<.01). Patients who received both norepinephrine and lidocaine drips (N = 14) were more likely to survive than those who received neither (N = 71; 57.1 percent vs 12.7 percent; p<.01). In contrast, patients who received a pacemaker (N = 60) were less likely to survive than those who did not (13.3 percent vs 33.9 percent; p<.01). A saturated logistic regression model revealed that all interactions between these three variables were nonsignificant. The comparison between a model including the main effects of these three variables and a model containing the main effects of all therapeutic interventions was nonsignificant, indicating that these three variables, taken together, provided an adequate fit of the data for the prediction of 24-h outcome.