Linear extrapolation of missing radiographic change scores in clinical trials does not spuriously overestimate group radiographic changes in rheumatoid arthritis.

OBJECTIVE: To assess linear extrapolation (LE) and last observation carried forward (LOCF) as imputation methods for radiographic change in patients with RA.

METHODS: The OSKIRA-1 trial enrolled 918 patients with active RA for studying the efficacy of fostamatinib. Radiographs were scheduled for all patients at baseline and week 12, regardless of early escape, and at weeks 24 and 52 for patients who remained in the study. Complete radiographic data for the 24-week follow-up were available for 623 patients and were assessed according to the Sharp/van der Heijde score. From this complete set of data, a random selection of 10% missingness was generated. This was done 1000 times, and for each replicate the missing radiographic change at week 24 was imputed, first by LE, then by LOCF. The mean of the mean and mean of the s.d. across the 1000 replications was calculated. A similar approach was iterated for different proportions of missingness.

RESULTS: The mean (s.d.) observed Sharp/van der Heijde score change from baseline to week 24 was 0.36 (2.39). With LE, the mean (s.d.) change was estimated as 0.36 (2.65), 0.35 (2.88), 0.35 (3.17), 0.34 (3.57) and 0.32 (4.45) with 10/20/30/50/90% missingness, respectively. With LOCF, the mean (s.d.) change was estimated as 0.34 (2.39), 0.32 (2.38), 0.30 (2.37), 0.26 (2.36) and 0.18 (2.34) with 10/20/30/50/90% missingness, respectively.

CONCLUSION: LE led to stable estimates of progression at the group level, but increasing variability with increasing proportions of missingness. In contrast, LOCF imputation systemically underestimated mean progression with increasing proportions of missingness, with artificially reduced variability estimates.