以就診為單位之門診藥物推薦

Abstract

門診藥物推薦系統可作為門診醫師開立處方的參考，亦有助於藥物整合的進行。我們使用台北市立聯合醫院的門診就診資料，並提出兩種模型，一種是以相互加強原理拆解藥物對於診斷的機率分佈，一種是以自注意力機制為基礎的神經網路模型建構診斷與藥物組合的嵌入向量，進行以就診為單位的藥物推薦。

Medication recommendation system can assist doctors making prescription, and it is also useful for medication reconciliation. We use outpatient data from Taipei City Hospital, and propose two approaches. One use mutual reinforcement to decompose distribution of medications, and the other one is building embeddings based on self-attention mechanism. We use these two methods to make encounter-based outpatient medication recommendation.