Background Radiography is an inexpensive, reliable and reproducible method to detect and quantify progression of damage, an important outcome in rheumatoid arthritis (RA) trials and observational studies. However, the distribution of progression scores is skewed with many low or zero scores. Analysis of such data is challenging, and the choice of analysis technique may influence the result. Current analysis practice is unknown.

Objectives We systematically searched the literature to identify current practice for the analysis of radiographic progression in clinical trials and observational studies of RA.

Methods PubMed Embase and Cochrane databases were searched (2006–2016) to identify studies that described analysis techniques to compare radiographic progression in at least two groups. Studies in animals, children as well as conference abstracts and studies not written in English were excluded. Titles and abstracts were screened by one researcher (SM); a second investigator (LvT) evaluated the included cases, doubtful cases and a random sample of the excluded cases.

Information on study design, sample size, assessment methods and analysis technique was extracted by one researcher (SM), in consultation with 3 others (LvT; MB and JT).

Results Of 5980 identified papers, 252 were eligible. 226 of these reports were on a single study while 26 were on multiple studies in one paper.

Of the 226 studies, 75 studies used parametric techniques, such as t-tests, ANOVA or linear regression to analyze the data. Of these only 12% took the skewed distribution into account. In 78 studies, the continuous data was categorized into two or more groups and analyzed with binomial or ordinal methods, such a chi-square tests or logistic regression analyses. 4 studies treated the outcome as a “count” outcome variable (2 studies applied a Poisson regression, 1 a negative binomial regression and 1 a zero-inflated binomial regression).

43% compared more than two groups. Median (IQR) sample size was 351 (range 163–608). 30% had one reader, 57% two readers, and 2% more than two; in 10% the number of readers was not recorded. Order of reading was random in 43%, sequential in 31%, and unknown in 26%. Most applied the Sharp van der Heijde scoring method (75%); 9% the Genant modification, and 16% other methods.

Conclusions There is large heterogeneity in the analysis strategy of radiographic progression in recent RA clinical trials and observational studies: a large number of studies apply simple, suboptimal or inappropriate methods. In addition, key information (i.e. number of readers and order of readings) is poorly documented.

Disclosure of Interest S. Mahmood: None declared, L. van Tuyl: None declared, L. Schoonmade: None declared, R. Landewé Employee of: director of Rheumatology Consultancy BV, D. van der Heijde Employee of: director of Rheumatology Consultancy BV, J. Twisk: None declared, M. Boers: None declared