一年追回超200億， 打擊醫保欺詐騙保行爲，有了AI新“武器”

{"type":"doc","content":[{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"醫保智能監控正在成爲打擊欺詐騙保戰役中的有力“武器”。"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"AI 大數據風控：打擊欺詐騙保戰役的“利劍”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"3 月 7 日，全國政協委員，濟南市副市長王桂英在其一份名爲“關於推進‘互聯網 +’醫保服務提升慢性病管理效能的建議”的提案中建議，加快推動國家醫保信息平臺建設，同時強化政策指導，完善醫保支付模式。此外還要強化監管效能，構建智能監控體系。加快完善統一的智能監控系統，通過生物認證、大數據風控、進銷存監管等手段，實現事前提醒、事中預警、事後審覈全鏈條監管，有效規範互聯網慢病服務醫療行爲，維護醫保基金安全。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"據官方披露數據，2020 年全國醫保基金收入 2.4 萬億，支出 2.1 萬億，收支平衡壓力不斷增大。另一方面，欺詐騙保行爲持續高發頻發。有數據統計，2020 年共檢查定點醫藥機構 62.9 萬家，佔全國所有定點醫藥機構數量的 99.8%，處理違法違規醫藥機構 39 萬家，追回醫保資金 219.2 億元。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但由於傳統的醫保風控主要是基於現行制度的規則引擎，如限定使用人羣、限定診療範圍、限定報銷上限等，制度標準通常只能從一個視角描述當前醫療行爲的適用範圍，不能對有計劃有目的的欺詐醫療欺詐行爲起到很好的攔截作用。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因此，加強醫保基金的監管，必須藉助信息化、智能化的力量。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"作爲平安集團旗下大醫療健康生態成員之一，平安醫保科技“大數據 AI 風控系統”研發團隊負責人在接受 InfoQ 採訪時表示，智能風控與傳統風控最大的區別在於對數據的利用效率上，基於大數據技術和 AI 算法，可高效地以多種時間尺度和空間尺度對海量數據進行定量的風險評估和篩選。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"醫保智能監控正在成爲打擊欺詐騙保戰役中的重要“武器”。據悉，平安醫保科技研發了“AI 大數據風控系統”，爲各省 \/ 市醫保局提供醫保風控的大數據採集、存儲、處理、分析和服務的整體解決方案。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"騙保行爲難逃 AI 算法“火眼金睛”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"據介紹，該司基於數據挖掘、自然語言處理、深度學習、圖計算等 AI 技術，結合平安的醫療、醫保行業經驗和知識庫，打造了具備“策略 + 模型”雙輪驅動引擎的“醫保鷹眼”系統，解決方案考慮到各地醫保局信息化成熟度和監管需求，從底層數據治理，到風控特徵構建，再到針對具體場景策略和模型，逐層完成了整個風控體系的框架搭建。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“大數據 AI 風控系統”架構基於 Lambda 架構，設計了 AI 大數據共享模型架構。系統在所有的運算過程分散在數據產生、計算和查詢過程當中，以統一的數據模型貫穿始終，從而提高整體的預算效率，滿足即時計算的需求。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"該負責人告訴 InfoQ，平安醫保科技 AI 大數據風控系統在構建的過程中，克服了不少技術挑戰。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"醫保支付風控具有事後、反饋週期長、業務耦合度深等特點，目前處於初步建設和方法論探索階段，對底層數據質量依賴性強，是我國醫保治理中的一大難題。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外，政府端的數據基建還不成熟，數據管制嚴格，很難建立針對醫保支付風控場景的完整的研發框架並長期對醫保支付數據特性進行研究。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“風控是業務屬性極強的數據挖掘場景，受政策、事件影響極強，各地市數據分佈、質量都有較大的差異，因此模型不用真實數據優化，無法達到理想的效果。所以要解決的實際問題是，如何在一個沒有標註數據的地方醫保局，建立、運行、並逐步優化支付風控系統”，該負責人表示。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"據瞭解，平安醫保科技基於遷移學習技術和專家經驗，實現了在無標註數據的情況下，建立並冷啓動風控模型。在減輕對數據依賴的同時，建立了準確的風控模型。隨着系統運行後逐步積累標註數據，模型會定期優化更新，和當地數據的適配度會越來越好。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"醫保領域的數據對用戶隱私要求高，數據越多模型識別越準確，但風控往往會與個人的隱私保護相沖突。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在隱私保護方面，平安醫保科技該負責人介紹到，公司針對各地市信息安全管理要求，並結合業務目標複雜度，採用了上雲、遠程、數據加密計算等技術方案 ，確保個人身份信息不參與到建模中，支付明細和診斷數據不出政府環境，甚至在不使用明文數據的前提下，完成風控模型的本地實施和優化。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“在數據標準化方面，還引入人工智能技術，對數據進行提取和結構化，再通過智能分類和映射等算法，配合少量人工覈驗，高效實現數據治理工作”。值得一提的是，數據標準化問題受到了兩會代表的關注。全國人大代表、湖南大學風險管理與保險精算研究所所長張琳提出加速醫療數據標準化，建立醫保數據與商業保險的數據共享機制，合理控費。建立數據安全使用標準以及數據共享機制能夠加快醫療系統信息化建設進度，讓數據在更加安全的範圍內被使用。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"無處遁形！360 度精準智能監管"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“欺詐與反欺詐，表面上是技術的博弈，本質上其實是人性善惡的博弈，有利益的地方就有違法犯罪動力”，該負責人說，“醫保基金體量巨大，信息化系統還在建設完善中，無法支持精細化、數字化、實時化的支付風險監管，勢必就有人動歪腦筋，鑽政策和法律的空子，從中套利 ”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"這也是爲什麼騙保行爲屢禁不止的根源。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"且近年來，新型欺詐騙保手段層出不窮，騙保違規行爲越來越隱蔽。例如，一些違法分子常採用“成羣結隊門診”的騙保方法。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"這一般是幾個或者十幾個參保人（醫保卡）多次發生的羣體行爲，主要是爲了開取市面上易銷售的藥品進行倒賣，成員往往具備一定的反偵查技巧，如自己的確患有某種慢性疾病，單次判斷所開藥品數量、種類等均符合要求，不違規。他們也很少會短期頻繁去一個醫院開藥，從頻次上也不會異常。這樣從藥品適應症、藥量、頻次等一般的監控規則上均難以發現違規。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"這樣隱蔽的違規行爲就可以通過“大數據 AI 風控系統”來識別發現。從算法角度，首先會利用各種關聯特徵（騙保分子可能與藥販子認識，騙保分子可能互相認識等）建立患者與患者、患者與醫療機構的同構和異構關係網絡，再利用稠密子集檢測算法抽取異常羣體，協同醫療稽覈崗最終認定出欺詐羣體。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"平安醫保科技 AI 大數據風控系統的智能監控針對全流程、全環節，其服務於醫療服務事前、事中和事後全過程，實施 360 度精準智能監管，發現隱匿的疑似違規案例，識別欺詐騙保行爲。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"平安醫保科技相關負責人表示，“不同於以往從業務經驗開始，找醫保騙保、作弊、違規等行爲的業務驅動模式，平安‘AI 大數據風控系統’是從數據入手，利用 AI 技術識別數據異常，在數據異常中識別騙保、作弊、違規等行爲。其中，AI 支持的異常識別，AI 支持的數據治理，對碼工作的自動化，表對齊的自動化等是該系統的核心競爭力”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"除打擊欺詐騙保行爲外，平安醫保科技的風控系統針對醫保項目的浪費、違規、濫用等行爲也建立了相應的監控模型，如對濫用、浪費等行爲設定的不合理檢查、不合理治療、不合理用藥等模型，通過大數據分析，尋找疾病的治療、檢查、用藥的常見行爲特徵，然後針對明顯超過常見行爲特徵的行爲進行監控分析，發掘異常之處。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"當下 AI 大數據風控技術在實際應用於醫保監管的過程中，存在一些待解決的難點，如風控模型可解釋性較差，缺少從模型的輸出到業務的使用的解決方案，如何指導醫保局在資源限制的條件下，最大程度的利用模型結果提升業務效果是一個難點。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外，還存在醫保風控效果難量化的問題。最重要的一個評估指標是——業務目標是否達成，即檢出欺詐案件數和案件金額。此外，還有時效性、複審準確率等指標。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“風控效果評估需要算法、醫療、稽覈、醫保局工作人員共同協作進行驗證。不過現階段，業內風控效果的評估方法、流程、標準還未建立起來。” 該負責人向 InfoQ 表示，平安醫保科技希望通過在重點地市與醫保局進行深入合作，來驗證技術的價值，基於業務需求定義出一套可量化的評價標準。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"結語"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"打擊欺詐騙保是國家醫保局成立後開展的一項重頭工作，自 2018 年以來已在全國範圍內開展多次打擊欺詐騙取醫療保障基金專向行動。在頂層設計上，多項政策大力支持醫保信息化建設，提倡推進醫保智能監控。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"去年 10 月，國家醫療保障局醫保信息平臺項目主體建設完成。目前，全國各地正在推進醫保信息平臺建設，預計 2021 年底在全國範圍內投入使用。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"伴隨着各地醫保信息化平臺落成，醫保智能監控水平也會再上新臺階。醫保信息化廠商將幫助醫保監管部門打擊欺詐騙保行爲，促進醫保資金合理利用。可以預見的是，作爲不可或缺的第三方力量，醫保信息化廠商未來將在醫保基金監管中發揮越來越重要的作用。"}]}]}