REGRESSION ANALYSES
Reading a Muitivariable Model. There are sev- eral key factors to isolate when a multivariable model is presented. First, how many variables were placed in the model in relation to the number of events (rule: there should be one variable per 10 outcomes)? Second, how many patients are in the study population? If the population is small, then the results may not be stable and generalizable to other patient cohorts. Third, were any major predictors left out of the model that could affect the outcome? Fourth, if they included the odds or relative risk ratios or even the coefficients, what is the width of the confidence intervals or standard errors? The wider the bounds, the less precise the estimate. This effect may often be seen in variables that are infrequent. When variables predict well, they will be consistent predictors across many data sets.
The interpretation of the odds or relative risk ratios are on a logarithmic scale, such that an odds of 4.0 is twice as great as one of 2.0. An example involves a model in which the number of perfusion defects is included and the relative risk of death is 3.3. The interpretation of this statistic is that for every reversible defect a patient's risk of death increases 3.3-fold. Finally, if the confidence bounds of the relative risk or odds ratios encompass 1.0, then the variable is nonsignificant.
Interpreting Diagnostic and Prognostic Analy-se~. There are several key points to examine within the results of a predictive model, including: (1) probability values, (2) test statistic, (3) coefficients, and (4) standard error values. When a regression model is performed, two types of p values are given: one for the entire model and