Both ß-blockers and non-potassium-sparing diuretics are mainstay antihypertensive therapies for many patients. Placebo-controlled trials have shown clinically meaningful reductions in the rates of stroke, coronary heart disease, and mortality when hypertension is controlled by these agents [1, 2]. We should not, however, universally accept the utility of these drugs without considering potential untoward side effects. We must critically examine studies that attempt to show an increased risk for sudden death associated with these agents. It is under the scrutiny of critical analysis that we feel the recent article by Hoes and colleagues [3] has serious shortcomings.

Two main areas require additional discussion. The first concerns the comparability of patient groups. Although the authors thoroughly searched for confounding and comorbid diseases, the two patient groups are clearly dissimilar. In all areas of known risk for death from cardiovascular disease recorded in the medical record (for example, history of cardiovascular disease, cardiovascular risk indicators, and comorbid conditions), case-patients had a prevalence significantly greater than that of controls. The authors attempted to control for these measured differences using multivariate conditional logistic regression. However, it is likely that unmeasured confounders (resulting because physicians chose these patients for a given antihypertensive therapy) would strongly influence their results.

Our second and principal concern was the authors' choice of analytical strategy. We believe they erred in doing multiple post hoc analyses without adjusting the level of statistical significance. This adjustment is imperative for rejecting the null hypothesis that the rate of sudden death does not differ between exposed and unexposed patients [4]. This phenomenon is best shown in the crude relative risks shown in the authors' Table 2—all are insignificant. The authors were able to report marginally significant relative risk ratios only after making post hoc adjustments, but they retained a 95% CI to define statistical significance. A much more stringent level would be necessary to prevent a type I statistical error (for example, a 99% CI). Additionally, many methodologists believe that a cause-effect relation in case-control studies must be supported by a relative risk greater than 3.0 [5]. On the basis of these concerns, we believe the results of this study must be viewed cautiously.