OBJECTIVES: Stroke–one of the leading causes of death and disability–represents substantial clinical and economic burden. Understanding treatment patterns and causative relations may help e.g. to identify outcomes predictors and cost drivers. METHODS: We used Polish Hospital Stroke Registry data on patient baseline characteristics (demographics, risk factors, prestroke disability, stroke severity), hospital management, treatment outcomes and drugs (pre-admission, during hospitalisation, and prescribed at discharge). We used inferred causation approach that deducts causal (not associative) interpretations from patterns of (conditional) independence. In this primary analysis we used 5000 observations from 2007/2008 year, binary variables and assumed no hidden variables. We used Tetrad 4.3.10-6 with PC algorithm. Variables were grouped into five tiers, a priori forbidding some directions of causal influence. Large number of variables led us to a restrictive significance level (α=0.0001). RESULTS: New insight can be gained from existence, lack of, and the direction of causal relations. Our results: confirmed (without imposing prior knowledge) the use of cardiovascular drugs in relation with underlying risk factors and natural sequence of drug management (drugs used prior to, in acute stroke and at discharge); surprisingly suggested no causal relation between some clinical characteristics and drug use (e.g. history of stroke/diabetes and oral anticoagulants) or acute stroke treatment (e.g. aspirin, thrombolysis, stroke unit based treatment) and mortality/post-stroke disability; determined the causal direction between some risk factors (e.g hypertension and diabetes, gender and AF) or patient history and prestroke disability (history of stroke or age over 75 and impaired disability); could not unambiguously discover the causal-relation between stroke unit based treatment and unproven efficacy drugs use. CONCLUSIONS: Apart from statistical or econometric approach, causal-type reasoning can be used both to confirm the intuition, and to detect new patterns in data. Further research should include the possibility of hidden variables and try to quantify the causal relations.