We present the HASARD method that is an hybrid approach for extracting adaptative temporal association rules. This method extracts assocation rules between events ccuring in subsequent time-intervals using closed itemsets extraction and evolutionary techniques. An important feature is its capacity to consider different time-intervals depending on the analysed attribute. This method was applied for the analysis of long term medical observations of atherosclerosis risk factors for cardio-vascular diseases prevention. Experimental results show that it is well-suited for extracting knowledge from temporal data where interesting patterns have different observation period length.