Risk assessment and risk stratification in sudden cardiac death: A biostatistician's view

Determining individual probabilities of developing lethal arrhythmia over time (risk assessment) and grouping individuals by that probability (risk stratification) are similar to, yet differ in purpose from, screening, diagnosis, risk factor identification, and prognostic staging. Methods of handling bias, use of multiple predictors, and evaluation of results provide challenges. A key purpose of risk assessment and stratification is examined. The role of operational definitions of predictors and events and of methods that account for multiple predictors and known confounding factors is analyzed. Constructed examples illustrate potential pitfalls in assessment and how multivariate techniques can deal with multiple predictors. A trial design to evaluate risk stratification for the identified purpose is elaborated and potential results are interpreted. Bias from predictors regressing to the mean can be minimized either by averaging a number of measurements or by equalizing the bias in comparison groups. An analysis of two predictors and two risk strata illustrates how the discrimination of combined predictors may be greater than the sum of the individual variables' discrimination. Risk stratification can be evaluated in trials that randomize competing interventions within different risk strata. Results of such trials indicate whether the risk strata adequately distinguish individuals by their responsiveness to particular intervention. Potential pitfalls, not easily recognized in risk stratification, can be avoided in the methods and in studies for evaluating those methods. Multivariate techniques maximize the discrimination of multiple predictors, but may increase complexity. Randomized trials of treatment provide evidence for utility of risk stratification.