台灣糖尿病人接受A1c檢測長期變化分析

Abstract

糖尿病的臨床照護盡責度，是近年糖尿病照護品質中相當受重視的一環，糖尿病臨床照護盡責度項目，包括：A1c血糖檢測、血脂檢測、血壓檢測、眼睛檢查、足部檢查，腎病變檢查及戒煙諮詢。由於A1c血糖檢測對於糖尿病的評估最為直接且被大眾接受。故本研究以台灣糖尿病臨床照護指引所建議之A1c年度檢驗次數，一季一次的A1c檢驗作為分析指標，對A1c檢驗的盡責度變化進行分析。 台灣的糖尿病照護雖然均歸屬在全民健保的制度下，但是，對於不同的分局別，醫院評鑑等級的照護是否有其差異性存在，過去並不清楚。在國外已有糖尿病照護盡責度的相關研究報告指出，在不同的地區、承保單位、醫院以及不同的醫師都會造成糖尿病照護品質上的差異。因此，本研究探討台灣糖尿病人A1c檢測盡責度變化，以及探討校正糖尿病患其他影響因素之下，不同健保分局別、不同評鑑等級的醫療院所對於糖尿病人A1c檢測盡責度變化的影響。 方法：本研究是以2000年台灣糖尿病世代追蹤健保資料庫之糖尿病人為研究對象，共620,854人，追蹤2001年到2005年五年間所有健保檔的門、住診資料。依照五年間糖尿病人每兩季之A1c檢驗狀況，將其分為四個型態類別，(1)連續兩季接受A1c檢驗之病患，(2)兩季皆無接受A1c檢驗之病患，(3)前一季有接受檢驗，而第二季沒有接受之病患，(4) 前一季沒有接受檢驗，而第二季有接受之病患。上述四種分類，分別以邏輯斯迴歸以及廣義線性混合模型，就研究對象的(1)人口學變相：性別、年齡，(2)醫療提供者面向：就醫分局別、就醫評鑑等級，(3)時間面向：就醫年度別，(4)病人特徵面向：Adjusted Clinical Group case-mixed system(ACG)病人疾病嚴重程度、糖尿病用藥以及用藥遵囑性，探討上述因子對糖尿病人A1c檢驗盡責度變化的影響。並藉由二元馬可夫鏈模式，以A1c檢驗機率轉移之兩條邏輯斯迴歸模型，就不同分局別、醫療院所評鑑等級別的A1c檢驗盡責度分佈差異進行分析，進而推估長時間追蹤的A1c檢驗穩定平衡機率。 結果：2001年到2005年五年間的糖尿病A1c盡責度檢驗完成率由28%逐年提升到55%以上。糖尿病檢驗盡責度方面，連續接受A1c檢驗的糖尿病人由2001年16.65%提升到2005年的22.04%，連續兩季沒有接受A1c檢測的盡責度變化也由51.96%下降至43.84%。在校正其他因子後，A1c檢驗盡責度變化在各因子的影響方面，A1c檢驗隨著年度呈現逐漸改善的趨勢，人口學面向：女性優於男性、年輕的族群較易接受A1c檢測，醫療提供者面向：就醫分局別有地區性，台北分局、高屏分局與北區分局的A1c接受度與連續檢驗的比率較佳，中區分局、南區分局、東區分局則較差，就醫醫院的評鑑等級以醫學中心最佳，依序區域醫院、地區醫院，以診所最差。病人特徵面向：病人疾病嚴重度以症狀輕微、中度、重度病患為最佳，非常嚴重以及健康病患則較差。降血糖口服用藥以及胰島素，都是以有使用者較無使用藥物的病患，在A1c檢驗表現較佳。用藥遵囑性是無用藥族群的A1c檢驗表現最佳、用藥遵囑性好為其次，遵囑性差的表現為最差。 結論：A1c盡責度在台灣整體的醫療環境是呈現大幅的正向成長，但是在醫療提供者面向的影響依舊呈現一個穩定方向性的差異，在醫療院所評鑑等級方面，評鑑等級越高的醫療院對於A1c檢測盡責度照護越佳，相對的，在診所的A1c檢測盡責度則需要加強。此外，A1c檢驗盡責度在分局別之間亦有很大的差異，台北分局與高屏分局在A1c檢驗的表現最佳，南區與東區最差，可見A1c檢驗有地域性的差別。以二元馬可夫鏈，探討A1c檢驗在不同分局別與醫療院所的轉移機率，發現同於羅邏輯斯回歸以及廣義線性混合模型的結果，但在未來針對醫療提供者之糖尿病照護改善介入的建議，除了醫院層級、各分局的差異，建議針對未檢驗轉移到檢驗的族群，以期獲得較大的成效。 研究限制：本研究礙於資料型態與資料保護協定，導致無法取得A1c生化數值，並有許多自變項的因子無法納入分析，例如：教育程度、研究對象年收入、看診醫師、抽煙與否、BMI等變項。又因研究設計，而無法得知每一位研究對象實際的罹病時間，故不將罹病時間長短納入研究。阿斯匹靈用藥的部份，考慮到台灣醫療給付的緣故，阿斯匹靈用藥在糖尿病病人的處置可能是以不同的疾病編碼歸類，若以全年度的用藥考慮，則會有over estimate的問題，故對於阿斯匹靈的用藥因子也不予與採納。這些未被採用的變項可能會對影響並人接受A1c檢測的行為產生影響而，對本研究產生誤差。本研究雖然使是經過校正的糖尿病病人資料，但在醫院因應健保給付制度的情形下，健保資料仍會存有一定的誤差，健保檔的ICD_9_CM的錯誤編碼會直接導致、ACG疾病嚴重程度別都會受到錯誤編碼的影響，更可能導致研究結果的誤差。另一方面，由於研究樣本數太大導致所有變項皆達顯著水準，以及重複測量分析程式的對資料量的限制。本研究日後必須針對不同的分析方法，估計對應方法所需樣本數，使重覆測量估計能達所需樣本，也能在一般的邏輯斯迴歸中以最小樣本進行重複模擬，以期在日後對其資料的正確性做更進一步的校正。

The Diabetes clinical care performance attaches great importance to a part of the DM quality care in recently.thats including:A1c test, blood lipid test,blood preasure test,eye test,foot test, kidney disease examine and the consultation quit of smoking. A1c test is the most immediate evaluation and the most acceptive among above, and Therefore this study bases on the A1c annual testing frequency of Taiwan’s DM clinical caer guildline, Three months one time examine, to be our analysis indicator.And we aim to analyze Long-term Dynamic Change of Diabetes Patient care-A1c testing in Taiwan. Although The DM care all belong to one National Health Insurance, is ther existing difference between each branches. That shows not clearly in past. But there are some published reference point out that the DM care quality shows difference in various area, insurance, hospital, even doctor. So we want to analyze the NHI branches’ effect and the different hospitals’ effect of Taiwan DM patient’s A1c test accountability after adjuested other co-variate. Method: This study use the Taiwan DM cohort database(620,854 people, follow 5 years)as our study population. Accounting to this five year’s NHI data to collect each 2 season to divid four types by each patient(1) both season took A1c test. (2) first season took A1c test but second did not.(3) first season did not take A1c test but second did. (1) both season did not take A1c test.we use logistic regression and GLIM model to measure the effect after adjusted below.(1)demography variate:sex、age (2)medical provider:NHI branch、hospital level (3)patient characteristics: Adjusted Clinical Group case-mixed system (ACG)、DM oral drug compliance and DM drug (4)time:different year.Finally we use monte-carlo simulation method to modify long-term charged probability in Taiwan DM patient’s A1c test. Result: Taiwan DM patients’ A1c test achievement percent grow up from 28% to 55% in five years.and type 1(both season took A1c test) becom 16.65% to 22.04%. type 2(both season did not take A1c test) becom 51.96% to 43.84%.after adjusted other covariate, A1c accountability shows improvement as year’s passing.Demography dimensoion:weman’s performane beter than men,young people better than old people. Medical provider dimensoion: medical center is the optima, area hospital the next,and clinic is the worst. And Taipei branch, Kao-Ping branch and northern region are better than other three branches.Patient characteristics dimension:the ACG shows more healthier patient have better performance in A1c test.as same result in Drug and compliance. Conclusion:Although A1c test keeps growing up in Taiwan,the medical provider dimension effect still present a stable difference.our study result responses to there is a area difference in DM care qulity of before reference.On the other hand,the markov chain’s trans-probability between hospitals shows the same thing as logistic and GLIM model shows.we suggest authority should address type 2(both season did not take A1c test) to gain more benefit. Discussion: This study was impeded by NHI data can not offer personal information.so that we need to exclude many variate to analyze, ex: education、personal income、visit doctors、smoke or not、illness period and BMI. In addtion, we also put debatable drugs in NHI payment away,like aspirin. We can’t clearly defined which visit prescription is only for DM.if we calculate all. There will be over-estimate problem.Beside drugs, NHI’s ICD_9 coding as usual have some problem in ACG system.Further more, in order to solve the restriction on GLIM model in SAS 9.2 edition and logistic regression evaluation. Sample size estimate should be considered in future.