We performed a sensitivity analysis to determine the robustness ofeffect estimates against within-study bias risk. Sensitivity analyseswere conducted by excluding sequentially one study and adding subgroups:(1) studies with a short versus long follow-up (≥4 weeks versus<4 weeks), (2) ischemic stroke versus intracranial hemorrhage,(3) sample size (n≥4000 versus n<4000), and (4) type of coronaryartery disease (stable versus acute coronary syndrome). Six variableswere included for assessment of bias risk: adequate sequence generation,allocation concealment, blinding to treatment arm, blinded treatmentassessment, intention-to-treat analysis, and incomplete outcomedata assessment. Trials without blinded outcome assessment or blindingto treatment arm were assumed as studies with high bias risk. Inaddition, numbers needed to treat were computed using the inverseof underlying risk as indicated by the control event rate. Small studybias or reporting bias was appraised by graphic inspection of funnelplots. Unadjusted P values are reported throughout, with hypothesistesting set at the 2-tailed significance level of 0.05. Variables are reportedas number or percentage as appropriate.
การแปล กรุณารอสักครู่..