Descriptive statistics summarize the characteristics of the patient sample. A multinomial logistic regression model was used to estimate probabilities for the 3 nominal response categories of attempted resuscitation, no resuscitation, and withheld resuscitation. Independent variables included factors of time (preguideline or postguideline implementation), participating agency (yes or no), patient location (private residence, nursing home, or other), patient sex (male or female), a covariate of patient age in years divided by 100, and an interaction term of time and participating agency. Individual agency identification was not tabulated for each observation or medical incident report; therefore, a random-effects term for agency was not entered into the model. A multiple imputation method using the expectation-maximization algorithm for continuous and categorical data was used for missing data in 89 cases of age and 46 cases of sex. Chi-square tests compared resuscitation wishes (verbal or written) with study group and patient location. Summary statistics were performed with InStat for Macintosh, version 3.0 (GraphPad Software, San Diego, California). Multiple imputations, regression models, and chi-square tests were evaluated by using R, version 2.2.0 (R Core Development Team, 2005) (3).