Using large-scale observational data and machine learning, thrombolysis, in real world use, was found to have at least as much benefit as predicted by the thrombolysis clinical trial meta-analysis. Both qualitative research and machine learning revealed significant between-hospital variation in which patients receive thrombolysis, which is leading to significant between-hospital variation in thrombolysis use and outcomes. Machine learning revealed that who will benefit from thrombolysis is patient-specific, and not easily captured in a simple medicine use label, but we found overall that stroke teams with a higher willingness to use thrombolysis are predicted to be generating better patient outcomes at a population level.