The MANTRA II study.

We applaud Mitchell Krucoff and colleagues (July 16, p 211) and The Lancet editors for publishing an extensive, well controlled, multisite study on prayer and healing. The Article and accompanying Editorial note that prayer had no significant effect on outcome in 748 heart patients. Study of the intangible is challenging, and the researchers should be recognised for their insight and courage. However, the results and discussion sections omit one point that could substantially alter the conclusions. The problem derives from the use of inferential statistics and hypothesis testing. In experimental studies, outcome measures between groups are compared; if the differences are significant, the null hypothesis is rejected and the alternative hypothesis tentatively accepted: the intervention is effective. The researcher must statistically test whether the results could have been obtained by chance. The value is usually set at 0·05: could this result occur by chance less than five out of every 100 tries? In the prayer study, the null hypothesis would be: “prayer has no effect on healing”. If significant differences are found, we reject the null. But what if the difference is not significant? The null hypothesis has not been shown to be true. The researcher should fail to reject the null, and conclude that all possibilities remain. But Krucoff and colleagues did something different: they accepted the null hypothesis, concluding that prayer had no significant effect. They omitted the conclusion: “therefore, all possibilities remain”. This study was not designed to show that prayer has no effect on healing in cardiac patients. The finding of significance required simultaneous alignment of numerous assumptions for an effect to be found. For example: is there a placebo or Hawthorne effect from participation in a prayer study? Did those praying know how to pray effectively, and receive enough information to connect the patient with a source of transcendence? Does researcher commitment to the study of prayer result in altered outcomes? Are there differences among types of prayer (eg, is emotional, heartfelt prayer overstimulating and detrimental to cardiac patients)? Is prayer so pervasive among cardiac staff and patients that a non-prayer condition is not discernable from a treatment condition? Do cardiac medical interventions have enough variability in outcome to measure such a subtle intervention? Is prayer clinically significant only within the outcome measure (eg, could intervention of a transcendent power result in death)? Lack of significant differences could have resulted from any of these, each of which might have confounded Krucoff and colleagues’ conclusions. Let us, as pioneers of science, continue to search for new ways to explore and understand the full gamut of human experience. Caution is indicated as we attempt to use the yardstick of statistics to measure the expression of infinity.