We have covered interpretation of risk stratification results to some degree in Chapters 1 and 4 thus far. We have seen that stratified data and logistic regression coefficients taken from one population can be used to standardize or to predict events in another. These are very powerful tools, and their proper use and interpretation require attention not only to the mechanics of the arithmetic, but also to the issues of bias and underlying population characteristics. In this chapter, we will review some of the familiar interpretation issues briefly, and then will cover some broader concepts related to the depth of the risk inquiry and the presentation of data.

Population characteristics and their distribution

The importance of comparability between standard populations and local study populations with regard to the distribution and prevalence of risk factors cannot be overstated. Multivariable risk models cannot reasonably compare populations that are grossly different with respect to their underlying risk characteristics. This is because the weighting of risk factor coefficients will be different in populations where risk factors are common than it will be in populations where the risk factors are rare. A risk factor that is rare in a population, even if it is strong, will account for only a small number of events in that population. If the events being studied are common, the small number of events attributable to the rare risk factor will also be a small proportion of the overall events, and the factor will not look important in that context.