How prior beliefs shape the interpretation of a randomized trial: a Bayesian re-examination of the CROSSFIRE trial

A randomized trial reports the probability of the data given a hypothesis; the reader needs the probability of benefit given the data, and that quantity cannot be computed without a prior. We re-read the CROSSFIRE trial of distal radial fracture fixation in older adults under five literature-anchored prior beliefs, ranging from strong equivalence to anatomic restoration, against two thresholds for meaningful benefit. At 3 months, defensible readers’ posterior probability of a clinically meaningful benefit spanned 0% to 47%; at 12 months, every prior converged on no meaningful difference. Every reader who moves from a trial report to a judgement about benefit brings a prior to that reading, whether or not it is stated. Disagreement among careful readers of the same trial can reflect different defensible priors rather than error, and making the prior explicit turns that disagreement into something colleagues and patients can inspect.