BACKGROUND: The risk of malignancies on TNF-α antagonists is controversial. The aim of this survey was to assess cancer risk on TNF-α antagonists in adult rheumatoid arthritis patients, including the five marketed drugs (infliximab, etanercept, adalimumab, golimumab and certolizumab) used in line with the New Drug Application. Furthermore, the relative interest of modified intention to treat or per protocol analyses to assess such sparse events remains unknown. METHODOLOGY/PRINCIPAL FINDINGS: Data sources were MEDLINE, CENTRAL, ISI Web of Science, ACR and EULAR meeting abstracts, scientific evaluation of the drugs leading to their marketing approval, and clinicaltrials.gov, until 31 December 2012.We selected double-blind randomized controlled trials in adult rheumatoid arthritis patients, including at least one treatment arm in line with New Drug Application. We performed random effect meta-analysis, with modified intention to treat and per protocol analyses. Thirty-three trials were included. There was no excess risk of malignancies on anti-TNF-α administered in line with New Drug Application in the per protocol model (OR, 0.93 95%CI[0.59-1.44]), as well as in the modified intention to treat model (OR, 1.27 95%CI[0.82-1.98]). There was a non-significant tendency for an excess non-melanoma skin cancer risk in both models (respectively, 1.37 [0.71-2.66] and 1.90 [0.98-3.67]). With fixed effect Peto model restricting to trials during at least 52 weeks, the overall cancer risk was respectively 1.60 [0.97-2.64] and 1.22 [0.72-2.08]. Whatever the model, modified intention to treat analysis led to higher estimations than per protocol analysis. The later may underestimate the treatment effect when assessing very sparse events and when many patients dropped out in placebo arms. In metaregression, there was no differential risk among the five drugs. CONCLUSIONS/SIGNIFICANCE: This study did not find any evidence for an excess cancer risk on TNF-α antagonists in adult rheumatoid arthritis patients, but an excess cancer risk after several years of exposure cannot be ruled out. Both modified intention to treat and per protocol analyses should be presented in such safety analyses.