Background/Purpose:

Intention-to-treat (ITT) principle is recommended to analyze randomized controlled trials (RCTs). It entails analyzing all subjects per the assigned group at randomization to avoid treatment effect estimation bias. Missing outcome data in RCTs can compromise results¹ and its wrong handling may cause bias. Single imputation methods e.g. “last observation carried forward (LOCF)” have unrealistic assumptions and discount the uncertainty of imputation². Sensitivity analyses are recommended to assess robustness of such assumptions ². We studied RCTs of rheumatoid arthritis (RA) for ITT use, missing outcome data handling, and sensitivity analysis usage. Trends between 2006 and 2016 were analyzed.

Methods:

MEDLINE and Cochrane Central Register of Controlled Trials database were searched to identify parallel-design, non-phase-1, English language, original RA RCTs of drug therapy with clinical primary outcome(s) published in 2006 and 2016. The study RCTs were assessed by two reviewers. ITT analysis was defined when all randomized subjects were included in the final analysis for the primary outcome(s). A modified ITT (mITT) allowed exclusion of one or more of the following: those found non-eligible after randomization, those who never received study intervention, and those without baseline or any post-baseline assessment ³. All other exclusions were classified as inadequate.

Results:

80 RCTs, 36 from 2006 and 44 from 2016, were eligible (Table 1). 48/69 (69.5%) RCTs had > 10% and 17/69 (24.6) had >20% missing data. RCTs doing ITT or mITT analyses enrolled more subjects (p < .001) but had similar completion rates (p = .734). Missing data of >10% was more in comparator arm vs. experimental intervention arm [16 (25.4%) vs. 3 (4.8%)]. RCTs performing ITT analyzed 100% subjects. Median (IQR) % analyzed for mITT RCTs was 98.8 (97.5-99.5) and inadequate analysis RCTs was 88.6 (81.7-92.1).LOCF was the most used imputation method. Utilization of ITT or mITT analysis from 2006 to 2016 was similar [23/34 (67.6%) vs 25/42 (59.5%), p = .465]. Only 42.9% and 57.1% RCTs reporting ITT analysis actually performed one in 2006 and 2016 respectively. Sensitivity analysis use between 2006 and 2016 was similarly low [6/36 (16.7%) vs 9/44 (20.5%), p = .666]. Only one 2006 RCT used the preferred imputation methods. Missing mechanisms and comparison of completers and dropouts were given by one RCT each.

Conclusion:

Missing data is common and most RCTs have >10 % missing data. While most patients are in the final analysis, many RCTs apply ITT principle incorrectly, utilize inappropriate single imputation methods and do not specify missing data handling methods. Sensitivity analyses and mechanism of missingness reporting is very low. No significant changes were seen from 2006 to 2016.

References:

1. O’Neill RT. Clin Pharmacol Ther. 2012;91:550-4
2. Little RJ. N Engl J Med 2012;367:1355-60
3. Dossing A. J Clin Epidemiol 2016;72:66-74

« Back to 2018 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/comparison-of-missing-data-reporting-and-handling-in-randomized-controlled-trials-of-rheumatoid-arthritis-drug-therapy-a-snapshot-ten-years-apart/