A COMPARISON OF MANUAL VERSUS COMPUTER-ASSISTED RADIOGRAPHIC MEASUREMENT - INTRAOBSERVER MEASUREMENT VARIABILITY FOR COBB ANGLES

Study Design. A comparison between computer-assisted measurement using digitized radiographs, which has the potential to reduce error, and m anual measurement using standard radiographs. Objective. To assess mea surement variability for the Cobb method on digital radiographs and co mpare it with that of manual measurements on standard radiographs. Bac kground Data. Studies of the Cobb method have demonstrated multiple so urces of error leading to significant intraobserver measurement variab ility. Estimates for the 95% confidence interval for intraobserver var iability range from 2.8 degrees to 10 degrees. Methods. Twenty-four sc oliosis radiographs were measured by six examiners. Two measurement se ts were done manually (''manual set''), and two measurement sets were done on digitized images using a computer mouse (''computer set''). Re sults. For the manual set, the 95% confidence interval for intraobserv er variability was 3.3 degrees (range, 2.5-4.5 degrees). For the compu ter set, the value was 2.6 degrees (range, 2.3-3.3 degrees). This diff erence in 95% confidence intervals between the manual and computer set s was statistically significant (P < 0.001). Conclusions. The results of this study demonstrate that intraobserver variability for manual an d computer Cobb angle measurements yield a 95% confidence interval of approximately 3 degrees, with the computer having a slightly lower var iability. The computer technique removes sources of intrinsic error, e .g., the variability introduced by using different manual protractors, the inaccuracy of standard protractors, and the use of wide-diameter radiographic markers. Identical digital images can be shared electroni cally between centers, without having to duplicate and mail films. Mul ticenter studies in which different examiners will be measuring Cobb a ngles may consider using the computer as a measuring device to reduce intrinsic measurement errors.