Skip to main content


Instrumental variable methods to adjust for treatment non-compliance and contamination in meta-analyses of randomized controlled trials

Date and Location




Sunday 22 September 2013 - 13:30 - 15:00


Presenting author and contact person

Presenting author

Benjamin Djulbegovic

Contact person

Branko Miladinovic
Benjamin Djulbegovic
Abstract text
Background: In systematic reviews of randomized controlled trials (RCTs), intention-to-treat (ITT) is the standard data analysis method which minimizes bias by including all patients regardless of the treatment received. Instrumental variable (IV) methods have been proposed as a technique to adjust ITT estimates for treatment non-compliance and contamination so that treatment effects in compliers may be estimated. Objective: To illustrate the value of IV methods by applying them to a published meta-analysis on the screening for breast cancer. Methods: We used ITT and IV meta-analytic pooled estimates of relative risk (RR) together with baseline risk to calculate number needed to treat to benefit (NNTB) one patient and their 95% confidence intervals. In the context of IV meta-analysis, NNTB can be interpreted as a number of assigned patients needed to comply to benefit one patient. Results: No heterogeneity was detected for ITT and IV pooled estimates for both non-compliance and contamination adjusted meta-analysis (I-sq = 0%, tau-sq = 0). The patient compliance rate ranged from 65% to 100% and contamination from 3.9% to 19%. The overall ITT estimate RR = 0.85 (95% CI: 0.75-0.96) translates to NNTB = 1,904 (95%CI: 929-6,378). Under non-compliance alone, for IV the overall RR = 0.81 (95% CI: 0.69-0.95), NNTB = 1,645 (95%CI: 980-4,348), which is 14% reduction from ITT estimate. Under non-compliance and contamination, for IV the overall RR = 0.78 (95% CI: 0.64-0.95), NNTB = 1,421 (95%CI: 806-4167), which is 25% reduction from the ITT estimate. Conclusions: Given that the U.S. Preventive Services Task Force’s recommendation against screening mammography in women aged 40-49 years were partly influenced by NNTB, the differences of 14% and 25% are not trivial. IV estimates appeal to patients who are eager to comply with their treatment assignment and to clinicians who wish to establish the efficacy of treatment received.