The CHARM program was designed as 3 separate randomized trials comparing candesartan with placebo in patients with chronic heart failure (CHF) who (1) were intolerant to angiotensin-converting enzyme inhibitor and had left ventricular ejection fraction (LVEF) < or =0.40; (2) were on angiotensin-converting enzyme inhibitor and had LVEF < or =0.40; or (3) had LVEF >0.40. CHARM provides an interesting example of the challenges faced by a Data and Safety Monitoring Committee (DSMC). Although the primary efficacy end point for each component trial was cardiovascular (CV) death or hospitalization for CHF, the primary outcome for the overall program was all-cause mortality. The DSMC received monthly safety reports and also met every 6 months (7 times in all) to review interim reports. Statistical stopping guidelines were predefined for all-cause mortality in the overall program. The overarching principle of the DSMC was proof beyond a reasonable doubt that would be likely to influence clinical practice. There were significant treatment differences in all-cause mortality for the overall program at several interim analyses, and the statistical stopping guideline was reached on one occasion. However, even a conventional level of statistical significance (P < .050) was achieved in only 1 of the 3 component trials. The DSMC consistently recommended that the program continue as planned. The final published result for all-cause death over a median of 3.1 years was a 9% reduction in hazard (95% CI 0%-17%, P = .055), whereas for CV death or hospitalization for CHF, there was a 16% reduction in hazard (95% CI 9%-23%, P < .0001). Subsequent exploratory analyses suggest that the hazard reduction in CV death was more marked in the first year after randomization and that, if real, this apparent treatment-time interaction offers a plausible explanation for why the interim mortality data showed statistically more extreme findings than the overall final results. The DSMC experience in the CHARM program illustrates the importance of continuing a trial to its scheduled completion unless there is proof beyond a reasonable doubt that would influence clinical practice rather than strict reliance on a statistical stopping guideline.