背景：與其他國家相比，台灣糖尿病的盛行率並不是很高，但是糖尿病死亡率相對高許多。過去研究也顯示國與國間譯碼員選擇原死因過程有相當大的差異。 目標：本研究以美國國立衛生統計中心發展的電腦化原死因選擇系統（ACME）希望瞭解台灣高糖尿病死亡率是否因為譯碼錯誤所造成。此外，也要檢討ACME選擇原死因決策表的可能問題，提出一個同時考率原死因理念與醫師死診填寫行為的糖尿病相關死亡分類架構，進而估計台灣糖尿病的真正死亡率。 方法：系統抽樣1994年5%死亡證明書5621份，本研究針對死亡證明書有提及糖尿病者408份進行分析。將死亡證明書上每個診斷都譯碼ICD-10代碼並輸入ACME軟體選擇原死因。首先比較譯碼員與ACME所推估之死亡率，接下來計算兩種方法的一致率、確認率與偵測率，再將不一致理由進一步分類。訪談五位譯碼員，了解不一致的理由及對ACME選擇之原死因的評論。綜合上述分析，研究者再提出一個糖尿病相關死亡的分類架構。 結果：在408份有提及糖尿病的死亡證明書中，有280份（68%）譯碼員選擇糖尿病作為原死因，有269份（66% ）ACME選擇糖尿病作為原死因。以兩種方法推估的糖尿病死亡率並沒有統計顯著差異。譯碼員與ACME不一致的理由：12%是譯碼員對疾病因果關係判定錯誤，88%是譯碼員喜歡選擇某些疾病（大多是心血管疾病）為原死因，這類不一致又可再細分為：1) 糖尿病在第一部份，較喜歡疾病在上一行（49/107）；2) 糖尿病在第一部份，較喜歡疾病在同一行（11/107）；3) 將填在第二部份糖尿病選為原死因（34/107）。訪談譯碼員表示許多選擇規則的解釋主要是顧問醫師的意見，她們也不太同意ACME選擇許多死亡機轉為原死因。 結論：台灣高糖尿病死亡率無法由譯碼員較喜歡選擇糖尿病為原死因來解釋，未來研究應該進一步探討是否是台灣醫師比其他國家較喜歡填糖尿病到死亡證明書上。本研究也發現譯碼員間的差異，主要來自醫師們對糖尿病在致死過程中角色認知之差異。ACME也過度選擇死亡機轉為原死因，本研究提出新的糖尿病死亡分類架構，特別將死亡機轉與醫師填寫死因診斷的行為因素考慮進去，相信應該更能滿足不同使用者之需要。 BACGROUND─Despite having a relatively low prevalence of diabetes compared with other countries, Taiwan has a relatively high diabetes mortality rate. Previous studies also revealed that the coding process among different countries sustained bias. OBJECTIVE─To determine whether the high diabetes mortality in Taiwan is due to errors in assigning diabetes as the underlying cause of death (UCD) by national coders, using Automatic Classification of Medical Entities (ACME) as the standard, and to examine implications for international comparisons. A framework of classification of diabetes-related deaths was proposed to estimate the real diabetes mortality in Taiwan. METHODS─Four hundred and seven death certificates in which diabetes was mentioned were analyzed. ACME was used to produce the correct assignment of the UCD, and sex- and age-specific diabetes death rates estimated according to national coders and ACME were compared. The national coders were then interviewed to determine their logic in UCD assignment and the reasons for discrepancies. The problems of ACME selected UCD were discussed and a framework of classification were proposed to estimate the real diabetes-related deaths. RESULTS─With ACME as the standard, more than one-fourth (107/408) of diabetes-related coding was in error. Nevertheless, owing to the compensation of opposite errors, the errors did not result in significant overestimation of diabetes mortality. Of the 107 discrepancies, 12% were due to ‘incorrect’ interpretation of causal sequence between diseases and 88% were due to the preference of the coder to select certain diseases, mainly cardiovascular diseases, as the UCD instead of diabetes. Interview results revealed that most of the discrepancies stemmed from 1) the opinions of consultant physicians on the role of diabetes in the dying process; and 2) controversies regarding the ACME decision tables. CONCLUSIONS─The high diabetes mortality in Taiwan is not due to coding errors; other factors should be explored. Remedies to reduce the frequency of discrepancies should not be confined to coders, but should also address consensus among physicians regarding ACME decision tables and the role of diabetes in the dying process. A new framework of classifying diabetes related death which considering both the mechanism of death and certification behavior was proposed to satisfy the needs of different users.
