Blinded reading of radiographs increases the frequency of errors in vertebral fracture detection

Abstract

We examined the effect of blinding X-rays to film sequence and patient identity on vertebral fracture detection. A sample of 50 postmenopausal women with low bone density and two sets of spine X-rays 3.6 years apart was selected; based on prior morphometric studies, about half of the women had experienced new fractures after the initial film. New morphometric and semiquantitative radiologist readings were each performed twice: blinded and unblinded. For morphometry, incident fractures were defined as vertebral height decreases of more than 15% compared with the initial film. The incidence was slightly higher when blinded versus unblinded (5.6 vs. 5.3% of all vertebrae for morphometry, and 9.7 vs. 8.7% for the radiologist), but these differences were not significant. The error rate was investigated by examining the frequency of "fracture reversals"-vertebrae identified as fractured on the initial film but not on the later film. Agreement between blinded and unblinded readings was generally greater than 80% for fractures but less than 10% for "fracture reversals," suggesting that reversals are not true events but random errors. The number of reversals was higher when the radiologist was blinded (2.1% of all vertebrae vs. 0.8% when unblinded; p = 0.07). The number of vertebrae with increases greater than 15% in size over time was also greater when morphometry readings were blinded versus unblinded: 0.8 versus 0% (p < 0.05). Although these errors are small, they are similar in magnitude to the annual fracture incidence in many populations. These data show that blinding X-rays to sequence offers no advantages, increases the frequency of errors, and may inflate incidence rates. We conclude that the assessment of X-rays for vertebral fractures in clinical trials should not be performed with the evaluator blinded to the sequence of the X-rays.