Background/Purpose: Intention-to-treat (ITT) method is recommended to analyze Randomized Controlled Trials (RCTs). It requires analyzing all participants per the assigned intervention group at randomization to prevent treatment effect estimation bias. In practice, RCTs have subjects with protocol deviations and/or missing outcomes. Incorrect analysis of these subjects may give biased results^1,2. Single imputation methods like “last observation carried forward (LOCF)” are discouraged as they may bring bias, and don’t adjust for the uncertainty of imputation³. We studied the ITT principle application, protocol deviations and missing outcome data handling in RCTs of rheumatoid arthritis (RA) drug therapy.

Methods: MEDLINE and Cochrane Central Register of Controlled Trials database were searched to identify parallel-design, English language, original RCT reports of drug therapy for RA with clinical primary and efficacy outcome(s) published in 2002-3, 2006-7 and 2010-11. Two reviewers assessed each RCT for the primary outcome(s) analysis and handling of missing data.

Results: 134 RCTs were eligible and enrolled a median (interquartile range, IQR) of 160 (59-381) patients. A median (IQR) of 83 (70-90) % remained enrolled until trial completion. 92 (67%) RCTs lost >10% to follow-up for the primary outcome. 84 (62.7%) reported using ITT analysis, 10 (7.5%) reported modified ITT analysis, and 40 (29.9%) mentioned no such term. 63 (47%) RCTs actually performed true ITT analysis (analyzed all patients as randomized, although 7 RCTs had missing data with undefined approach to imputation). 32 (23.9%) performed a modified ITT analysis [percent analyzed: 97 (95-99)%] with exclusion of following categories: those without post-baseline assessment (17); no intervention received (12), and non-eligibility (1). 20 RCTs analyzed only those completing the trial per the study protocol [percent analyzed: 89 (86-97)%], while for 19 RCTs patients exclusion reason was unclear [percent analyzed: 91 (69-93)%]. Only 5 RCTs had all subjects complete the trials. Only 92 of the 129 (71%) RCTs with a missing final outcome described missing primary outcome handling which included: LOCF (51), non-responder assignment to those with missing data (29), calculation by interpolation (5). Only 2 RCTs did sensitivity analyses with different missing data assumptions. Experimental intervention efficacy could be assessed for 124 RCTs. The percentage of positive RCTs (statistical significance favoring experiment intervention) according to analytic method were: ITT (66%), modified ITT (87%), per-protocol (67%) and unclear (50%) [p = 0.056, Likelihood Ratio test].

Conclusion: A considerable variation in reporting and application of ITT principle in the RCTs exists. In about 30% RCTs, per-protocol or unclear analytic methods were used. Similar number of RCTs with missing data did not address handling of such data. Vast majority of RCTs used the non-recommended single imputation methods for missing data. A better description of subjects selected for final statistical analysis and approaches to deal with missing outcomes are needed.

References:

1. Hollis S. BMJ 1999;319:670-74

2. Sedgwick P. BMJ 2015;350:h681

3. Little RJ. N Engl J Med 2012;367(14):1355-60

« Back to 2017 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/handling-of-missing-data-protocol-violation-and-performance-of-intention-to-treat-analysis-in-the-randomized-controlled-trials-of-drug-therapy-of-rheumatoid-arthritis/