Trend of salient characteristics over time in RA studies and comparison with a study from top four medical journals
| Characteristic | Time period of included studies/(reference) (publication year) | ||||
| 1994–200325 (2005) | 2006 (Current) | 2016 (Current) | 2013* 24 (2014) | 2008–2013* 27 (2016) | |
| Overall ITT analysis performed, % | 7.4 | 35.3 | 38.1 | 40.0 | NA† |
| Overall PP analysis performed, % | 59.3 | 11.7 | 28.5 | NA | NA |
| Preferred missing data handling methods used, % | 1.2 | 2.7 | 4.9 | 27.3 | 1.9 |
| Dropout in each study arm given, % | 69.0 | 82.3 | 85.7 | 97.0 | NA |
| Missing data handling given, % | 23.5 | 61.1 | 48.8 | 36.0 | 94.1 |
| LOCF used to handle missing data‡, % | 52.6 | 41.7 | 22.7 | 12.0 | 56.8 |
| Missing mechanism given, % | NA | 2.9 | 0.0 | NA | 7.8 |
| Sensitivity analysis performed, % | 11.1 | 13.9 | 20.5 | 37.0 | 27.0 |
| Comparison of completers/non-completers, % | 16.7 | 0.0 | 2.3 | 11.6 | NA |
*Studies involving top medical journals.
†86% stated that ITT was used but no information on those that actually performed ITT.
‡LOCF was the most common method used across all years for imputation except 2016 (Ibrahim et al) where NRI was most frequent (74.5%) but NRI and LOCF were used simultaneously as well; in the 1994–2003 and 2013 studies, complete case analysis (59.3% and 45.0%, respectively) was the most common method to handle missing data.
ITT, intention-to-treat; LOCF, last observation carried forward; NA, not available; PP, per-protocol; RA, rheumatoid arthritis.