INTRODUCTION
Missing outcome data (MOD) may undermine interpretation of RCT by weakening power and limiting apparent effect size. We assessed bias and inefficiency of two imputation methods commonly used in stroke trials and how these may affect apparent efficacy of treatment (OR).

METHODS
We searched VISTA-Acute for ischaemic stroke patients with 90-day mRS as an outcome, and known thrombolysis status. We excluded any with missing 30-day mRS. We planned two analyses. First, we calculated OR for outcome thrombolysis versus not thrombolysed from imputed-only data, a) among patients with missing mRS90, and b) among matched patients with intact data (using propensity score methods and relevant covariates). Imputation approaches were last observation carried forward (LOCF) or multiple imputation MI. Outcome comparisons used dichotomization and shift analysis. Thereafter, we calculated whole-population OR using LOCF and MI (also through dichotomization and shift analysis); first with the original 1.5% MOD, then artificially increasing the MOD burden (5%; 10%; 20%; 30%). 
RESULTS
We considered 9,657 patients, 3,034 (31%) thrombolysed. MOD replacement by LOCF with analysis by dichotomization gave a high estimate of thrombolysis influence (OR 1.72 [1.45-1.99] p=0.001). Imputing an increasing burden of MOD progressive raised the OR estimates, though thresholds for overestimation were 10% MOD burden for LOCF; 20% for MI. 

DISCUSSION
Replacing missing outcome data tended to overestimate outcome differences of thrombolysed versus non-thrombolysed patients, especially if last observation carried forward rather than multiple imputation and dichotomization rather than shift analysis was used; but had minimal impact below a 10% burden of missing data.

MISSING OUTCOME DATA MANAGEMENT IN ACUTE STROKE TRIALS TESTING IV THROMBOLYTICS. IS THERE RISK OF BIAS?