We do not believe, as implied by Dr. Horton, that the results of ISIS-4 or LIMIT-2 should be considered "correct" or "incorrect." Rather, as stated in our discussion of meta-analysis, these well-designed trials should be seen as similar—but not identical—experiments, the results of which differ and, therefore, need careful consideration. We chose the example of magnesium in patients who have had acute myocardial infarction because large trials with conflicting results are available and because we wished to show how different methods of meta-analysis can further lead to different interpretations of the same data. At a pathophysiologic level, we agree with Dr. Horton that issues of timing for magnesium therapy may be important and believe that this issue can only be resolved by further trials specifically addressing this concern [1].

We did not suggest, as Dr. Resch implies, that meta-analysis should not inform decision making. Particularly in the absence of adequate randomized, controlled trials, meta-analysis constitutes a critical part of the database of clinical judgment. As we stressed, however, that judgment must be made with knowledge of the inherent methodologic and statistical constraints of even the best meta-analyses.

For therapy in which estimated treatment effects are substantial, large-scale clinical trials are not required. When estimated risk reductions are modest to small, however, clinical trials of substantial size must be done if informative null findings are to be distinguished from truly positive results. Although this approach may require multicenter or multinational cooperation, such collaborations are likely to produce data that are well worth the additional effort.