Sentinel events (SEs) are defined as unintended healthcare events that cause significant injury or death. Once a situation is recognised as an SE, the process of learning the lessons of an SE has several distinct stages: 1) reporting the SE, 2) analysing the SE, 3) formulating recommendations, 4) implementing recommendations and finally, 5) evaluating the effect of implemented recommendations. Many countries have developed a system to register and analyse SEs in attempt to prevent recurrence and improve patient safety. However, this has not yet resulted in a decrease in the number of SEs. Recurrence of similar SEs is a better measure for the effect of learning from SEs. Despite previous analyses of these type of SEs, they still recur on a daily basis worldwide. Recurrence of similar SEs strongly suggests learning from AEs is complex and unsatisfactory. Completing the learning process and optimising the different stages is expected to lead to quality improvement and a decrease in the number of recurrent SEs. However, objective standards to assess the quality of recommendations are lacking. To help fill this gap, this thesis primarily addresses the second and third stages of the learning process as detailed above, namely optimisation of the quality of analysis and subsequent recommendations derived from SEs. We specifically focus on developing a strategy to determine if a recommendation will be effective, and if and how performance variability is recognised and included in SE analysis and subsequent recommendations.