Objectives To evaluate if the mean smallest detectable change (SDC) of multiple time intervals using the Bland & Altman (B&A) levels of agreement (LoA) method is an appropriate surrogate for the generalisability analysis method for estimating the overall SDC of radiological progression in rheumatoid arthritis (RA) trials. Secondly, to compare the SDC based on 95% LoA with the SDC based on 80% LoA, and to investigate the association between SDC and baseline damage and progression.

Methods Fifteen datasets from randomised controlled trials in RA were scored by 13 experienced readers as pairs according to the modified Sharp/van der Heijde method. The SDC using the 95% and 80% LoA and the generalisability methods was calculated.

Results 21 295 radiographic time points from 7643 patients were included. The mean (range) SDC for the LoA and the generalisability methods was 3.1 (2.3–4.3) and 3.2 (2.3–4.6) units, respectively. The mean±SD difference between the two methods was −0.13±0.28. The mean SDC including all intervals (n=31) was 3.0±0.7 for 95% LoA and 2.0±0.4 for 80% LoA. No relationship was observed between baseline damage and the SDC, whereas the SDC increased with increasing radiological progression.

Conclusions The mean of the interval SDCs obtained by the simple LoA method is a valid surrogate for the SDC obtained by complex generalisability methods. The SDC depends on the level of radiographic progression rather than on the level of absolute damage. In addition, the use of an SDC based on 80% rather than on 95% LoA is proposed.