Friedman's test comparable regression model

A paper for review in veterinary medicine looks at the impact of x-ray beam angle and the position of a horse’s hoof as determinants of radiographic diagnostic quality rated on an ordinal scale by a board certified radiologist. 26 limbs were randomly chosen and radiographed from 6 different angles with the limb in 6 different positions. The diagnostic quality was evaluated for all combinations of angles and positions using a Friedman’s test with post hoc analysis with Wilcoxon signed-rank tests and a Bonferroni correction. My question is , with an ordinal outcome could you fit a mixed effects proportional odds model with horse id as the random effect?

1 Like

Is the question of ‘diagnostic quality’ referenced to a specific type of finding? Or are motion artifacts, etc., part of the consideration? Instinctively, I doubt that any model you could ‘fit’ to these data would ever offer something better than a quick call to the radiologist, detailing the clinical signs and asking which views are most desirable. Also, what are the decision-theoretic aspects of this? Does the horse have to be sedated for these X-rays? If so, presumably you would want to obtain every view you could get while the horse was sedated. The question of which single view is best seems artificial to me.

Yes I agree. The study was done on cadaver bones from 26 disarticulated limbs placed in a hydraulic vice to mimic a normal horse limb. The radiograph of interest is in the foot but quality can depend on if the limb is positioned with toe elevated or not, and the angle of the primary beam. The authors placed the limb in 6 different positions and used 8 different angles on each limb. Therefore they had multiple comparisons (multiple p values). They used the Friedman test to compare all possible combinations of foot placement and x-ray angle. The outcome variable was an ordinal rating scale of diagnostic quality by two radiologists. I was asked to review the paper and the statistical analysis seems inadequate to me. Of course, I also wonder if the results are transportable. I think you are correct about the artificial nature of the paper.

1 Like

Apropos of which …

1 Like