[ 摘要 ]

背景：全民健保次級資料已成為醫療服務研究重要之資料來源，如何適當控制病人共病(comorbidity)情形之差異，為一重要議題，然國內尚未有研究針對不同學者發展之共病測量方法進行實證探討。

目的：利用全民健保資料庫，比較不同共病測量方法之相對表現及其適用性，並發展新的共病權重。

方法：採回溯性世代研究(retrospective cohort study)，選取5種共病測量方法進行比較，包括Deyo、D-M、D’Hoore三種版本之Charlson Comorbidity Index(CCI)、以次級資料發展之Elixhauser、及以藥物處方情形發展之CDS 。以2001年住院病人作為發展組，追蹤其一年存活情形，求得本研究各共病類別之權重；再以2002年因急性心肌梗塞、肺炎、第二型糖尿病伴隨併發症、慢性腎臟疾病、充血性心衰竭、慢性阻塞性肺病住院之6種病人為對象，比較不同共病測量方法預測院內及住院一年內死亡情形之差異。共病採用類別、原始權重及本研究權重方式處理，共病之判定分別採用當次住院、當次併前一年住院、當次併前一年住院及門診3種資料來源。控制變項包括年齡、性別、原住民身分、及是否手術。以邏輯斯迴歸之c統計量比較各方法之相對表現。

結果：類別模式下，大致皆以D-M’s CCI及Elixhauser方法表現較佳。權重模式時，本研究發展之CCI新權重較原始權重表現佳；且以D-M’s CCI表現較佳。對CCI方法，以當次併前一年住院資料預測院內及住院一年內死亡情形表現最佳。對Elixhauser類別方法，預測院內死亡情形，僅以當次住院資料表現較佳；增加住院前門診就醫資料則無法改進預測院內及住院一年內死亡情形之表現。

結論：本研究發展之CCI新權重優於原始權重，可供後續研究者參考。三種版本之CCI方法中，以D-M’s CCI表現較目前國內最常用之Deyo’s CCI為佳，建議未來研究者可選用此方法。當研究樣本數夠大，可選擇類別方式之D-M’s CCI或Elixhauser方法。若研究樣本數小，建議選擇權重模式之D-M’s CCI。



[ 英文摘要 ]

Background：Comorbidity is an important controlling factor in health services research using administrative data. Although National Health Insurance databases have become an important resource for studies in this field; there is no study to investigate the relative performance of various available claims-based comorbidity measures in Taiwan.

Objective：To compare the performance of different claims-based comorbidity measures and to develop the new weights using National Health Insurance databases.

Method：Five different comorbidity measures, including Deyo’s Charlson Comorbidity Index (CCI), D-M’s CCI, D’Hoore’s CCI, Elixhauser and Chronic Disease Score (CDS), were chosen for investigation in this retrospective cohort study. Empirical weights were derived from the inpatient data in 2001. Patients admitted to one of six medical categories (acute myocardial infarction, pneumonia, diabetes mellitus with complications, chronic renal disease, congestive heart failure, chronic obstructive pulmonary disease) in 2002 were selected to compare different comorbidity measures. The comorbidity measures were implemented as individual components (the presence or absence of the comorbidity), and also as an index (weighted sum of comorbidity indicators). The comorbidity measures were created based on 3 sources of data: the index hospitalization, the index and prior hospitalizations, and the index and prior hospitalizations as well as outpatient visits. Control variables included age, sex, race, and operation in hospital. The c-statistic of logistic regression was used to compare the performance for predicting 2 outcomes: in-hospital death and 1-year mortality. Results：Better discrimination was achieved with the D-M’s CCI or the Elixhauser method when using individual components. When comorbidity measures were used as indices, better discrimination was achieved with the D-M’s CCI. The empirically derived weights of CCI performed better than the original one. For CCI methods, patient information available from both the index and prior hospitalizations performed best. For Elixhauser method, patient information available from the index hospitalization performed best when predicting in-hospital death. Adding prior outpatient data didn’t improve the performance of measures.

Conclusion：The empirically derived weights of CCI performed better than the original one, and it could be considered to use in further researches. D-M’s CCI performed better then Deyo’s CCI, which was used most frequently in Taiwan, and further investigators may consider to choose this method. When sample size is large enough, D-M’s CCI or Elixhauser method could be chosen and implemented as individual components. If sample size is small, D-M’s CCI used as an index would be the better choice.