The Byrd Coronary Care Prayer Study

Randolph Byrd was a cardiologist at San Francisco General Hospital who randomised patients admitted to the coronary care unit into two groups, passing the first names and diagnoses of one group to intercessors elsewhere who prayed for them daily, and leaving the second group unprayed for. Neither the patients nor the hospital staff knew who was in which group. He then compared the two on a long list of clinical outcomes and on a composite score of his own construction that graded each patient's course as good, intermediate or bad. The prayed-for group came out better on that composite.

The paper appeared in the Southern Medical Journal in 1988 and became, for the next two decades, the single most cited piece of evidence in popular argument that prayer has measurable medical effects, quoted in books, sermons and news coverage far more often than it was read. Its actual claims were modest, and Byrd himself did not overstate them in print, but the paper's afterlife in secondhand citation was another matter entirely. The same background-prayer problem applies here and Byrd acknowledged it: patients in a coronary care unit generally have families, and no trial can arrange for a control group that nobody prays for. What a trial can do is compare more organised prayer against less, which is a narrower question than the one the study is usually cited as answering, and one worth keeping in view whenever any result in this literature is quoted.

The statistical objection to this trial is standard and is worth spelling out because it applies far beyond prayer research. If a study measures a couple of dozen outcomes and tests each at the conventional threshold, roughly one in twenty will separate the groups by chance alone even if nothing whatever is going on. Reporting only the ones that separated, or folding them into a composite constructed after the results are known, hides that arithmetic. The remedy is to name a single primary outcome before the data are collected and to correct the threshold for everything else examined.

Byrd's trial did not do this, and the outcomes that reached significance were not ones a stated prior hypothesis had identified. William Harris and colleagues attempted a stricter version in 1999 with a larger sample and reported a small positive effect on a similar composite, drawing similar objections. The question was then settled as far as this literature can settle anything by the STEP trial in 2006, which pre-specified its outcome, enrolled roughly four and a half times as many patients, and found nothing. A Cochrane review has since concluded the evidence does not warrant further trials.