In　this　article,　we　propose　a　pipeline　to　mine　interpretable　knowledge　from　electronic　health　records　(EHR)　for　the　Heart　Failure　(HF)　prognosis　risk　evaluation　task.　Mortality　risk　after　first-diagnosis　HF　highly　impacts　patients’　life　quality,　and　is　helpful　for　physicians　to　efficiently　monitor　patients’　disease　progress.　How　to　mine　medically　reasonable　and　interpretable　knowledge　to　assist　physicians　in　evaluating　mortality　risk　is　a　non-trivial　task.　The　proposed　pipeline　leverages　a　gradient-boosting-based　predictive　model　to　estimate　the　risk　of　HF　prognosis,　and　discovers　variables　and　decision　rules　from　the　predictive　model.　The　mined　knowledge　is　confirmed　as　interpretable　and　inspirable　by　physicians.　Copyright　©　2021　for　this　paper　by　its　authors.