Background: The intention to treat principle (ITT) should be the main approach for analysis of randomised controlled trials (RCTs). However, trials reporting the modified intention to treat (mITT) are increasingly appearing in the medical literature and their impact on systematic reviews and meta-analysis is unknown. Objectives: To explore if bias was associated when analysis was conducted by mITT in randomised controlled trials. Methods: We performed a computerized search to retrieve systematic reviews and meta-analyses in PubMed and The Cochrane Library between 2004 and 2008. To be considered, reviews had to perform a quantitative analysis and at least one RCT that performed an mITT had to be included. Data related to reporting of the type of analysis (ITT, mITT, or no ITT reporting), other methodological quality measures (sequence generation, allocation concealment and blinding), and number of excluded subjects were extracted. A meta meta-analysis was performed and ratio of odds ratios (ROR) was calculated. A ROR less than one implies that trials reporting mITT exaggerate intervention effect estimates. Results: Of 31 meta-analyses, 355 RCTs were analysed. One hundred fifty-nine trials reported mITT, 95 trials reported ITT, and 97 trials did not report any type of approach. Sixty percent of RCTs using mITT approach reported exclusion against 33% of trials reporting ITT (p < 0.001). With respect to RCTs reporting ITT, trials reporting mITT significantly exaggerated the effect estimate by 18% [ROR 0.82 (0.72 to 0.96]. Conclusions: In our sample of meta-analyses the number of trials reporting modified ITT was substantial. Our analysis suggests that the mITT approach in RCTs when included in meta-analysis may introduce bias. Modified intention to treat approach should be abandoned.