前言：利用健保資料庫發表的糖尿病相關研究論文愈來愈多，且使用的糖尿病個案審定程序並不一致。如果個案審定程序不準確，會導致研究對象錯誤分組，影響研究品質。然而，台灣關於糖尿病診斷編碼準確度研究很少，且沒有驗證各種個案審定程序的準確度研究。
研究目的：本研究包含兩個目的，一是想瞭解過去台灣使用健保資料庫之糖尿病相關研究論文中，使用的糖尿病個案審定程序情況。二是評估健保申報資料糖尿病國際疾病分類診斷碼準確度，找出哪一種個案審定程序可以獲得最高準確度。
研究方法：第一部份利用Pubmed搜尋利用健保資料庫的糖尿病相關研究論文共520篇，整理出每個研究使用的糖尿病個案審定程序。接著分析個案審定程序與年代、研究類型、期刊以及通訊作者的相關。第二部份隨機抽樣奇美醫療體系三個院區10,000人，抽樣條件為在2015年中至少有四次門診就醫紀錄的20歲以上民眾，分別抽樣永康院區6000人，柳營院區2500人，佳里院區1500人。本研究判定是否有糖尿病的金字標準條件：一是有糖尿病藥物處方紀錄者，二是血糖檢驗值異常者，三是電子病歷有醫師文字診斷糖尿病文字診斷。接著比對健保申報資料國際疾病分類診斷編碼與金字標準，並依照不同個案審定程序計算敏感度、特異度、陽性預測值、陰性預測值、kappa值、陽性概似比與陰性概似比。最後分析病人、醫院層級、科別以及醫師等特徵與不同準確度指標的相關。
研究結果：第一部份有關台灣使用健保申報資料發表的糖尿病研究論文，使用個案審定程序依照論文發表年代分析，有愈來愈嚴格的趨勢。尤其近幾年藥物相關研究越來越多，許多個案審定程序都包括有使用藥物的條件。第二部分有關健保申報資料國際疾病診斷編碼準確度分析之研究結果，個案審定程序要求門診出現糖尿病診斷編碼次數增加(越嚴格)，特異度及陽性預測值會提高，但是敏感度及陰性預測值會互補性下降。如果優先考量陽性預測值，個案審定程序〝兩次門診診斷〞，〝兩次門診診斷相隔30天〞，〝三次門診診斷〞，〝三次門診相隔30天〞的陽性預測值分別是：92.9%，93.8%，94.4%與94.6%。但是敏感度會逐漸下降，分別是：83.6%，81.6%，81.5%與80.8%。如果要提高敏感度，個案審定程序可以加上住院診斷資料，下列不同程序的敏感度與陽性預測值分別是：〝一次住院或兩次門診診斷〞89.7%與90.4%，〝一次住院或兩次門診診斷相隔30天〞88.9%與91.3%，〝一次住院或三次門診〞88.7%與91.8%，〝一次住院或三次門診相隔30天〞88.5%與91.9%。關於相關影響因素分析，男性病患比女性病患，高齡病患比低齡病患，就診內分泌科比其他科陽性預測值較高。
研究結論：台灣使用健保申報資料進行糖尿病相關研究的個案審定程序越來越嚴格。研究者可以依照研究目的選擇較佳個案審定程序，如果優先考量陽性預測值，本研究建議使用審定程序〝三次門診診斷(相隔30天)〞。

Background: More and more studies used International Classification of Diseases (ICD) codes to identify patients with diabetes. However, little is known on the validity of various ICD coding algorithms for identifying patients with diabetes using administrative data. Objectives: The first aime of this study was to describe frequency of different case definitions used in diabetes-related studies during the past two decades. The second aim of this study was to assess validity of various coding algorithms for diabetes cases. Methods: In the first part, we used PubMed to retrieve published diabetes-related studies using Taiwan using national health insurance claims data for analysis. We then analyzed the changes in frequency of various coding algorithms across years and associated factors. In the second part, we randomly sample 10,000 patients aged 20 years and above who have visited Chi-mei health system outpatient at least four times in 2015. The gold standard for patients with diabetes mellitus included antidiabetic drug prescription records, abnormal blood glucose test results and text diabetes diagnoses recorded in electronic medical records. We then compared the coding algorithms with gold standard to calculate different validity indicators (sensitivity, specificity, positive predictive value [PPV], and negative predictive value [NPV]). Thirdly, we examined possible factors associated with the performance of ICD coding. Results: In first part of this study, the algorithm used became more rigorous across years. More and more pharmacoepidemiological studies published which would add prescription of anti-diabetic drugs as part of criteria in case definition algorithm. In the second part of this study, the more number of diagnoses required in the case definition algorithm (i.e., more rigorous), the higher the specificity and PPV. However, the sensitivity and NPV would compensatory decreased. If we have higher priority on PPV, the PPV of algorithm “two outpatient diagnoses”, “two outpatient diagnoses separated at least 30 days”, “three outpatient diagnoses”, and “three outpatient diagnoses separated at least 30 days” was 92.9%, 93.8%, 94.4%, and 94.6%, respectively. The sensitivity would compensatory decreased, which was 83.6%, 81.6%, 81.5% and 80.8%, respectively. If we wished to increase the sensitivity, we could add criteria using inpatient data. The sensitivity and PPV was 89.7% and 90.4% for “one inpatient diagnosis or two outpatient diagnoses”, 88.9% and 91.3% for “one inpatient diagnosis or two outpatient diagnoses separated at least 30 days”, 88.7% and 91.8% for “one inpatient diagnosis or three outpatient diagnoses”, and 88.5% and 91.9% for “one inpatient diagnosis or two outpatient diagnoses separated at least 30 days”. With regard to associated factors: male patients, older patients, patients visiting metabolic specialists had higher PPV than their counterpart female patients, younger pateints and patients visiting other specialists. Conclusion: The coding algorithm used to identify patients with diabetes mellitus was more rigorous across years. Depending on the purpose of research (sensitivity or PPV), the researchers could select the best algorithm according to the information this study provided. If we have higher priority on PPV, we recommend that algorithm “three outpatient diagnoses (seprated at least 30 days) would have the best performance.

Amed S, Vanderloo SE, Metzger D, Collet JP, Reimer K, McCrea P, Johnson JA. Validation of diabetes case definitions using administrative claims data. Diabet Med, 28(4):424-427, 2011.
American Diabetes Association. 2. Classification and diagnosis of diabetes. Diabetes care, 40(Suppl 1):S11-S24, 2017.
Baldessarini RJ, Finklestein S, Arana GW. The predictive power of diagnostic tests and the effect of prevalence of illness. Arch Gen Psychiat, 40(5):569-573, 1983.
Barber C, Lacaille D, Fortin PR. Systematic review of validation studies of the use of administrative data to identify serious infections. Arthritis Care Res, 65(8):1343-1357, 2013.
Chan WC, Papaconstantinou D, Lee M, Telfer K, Jo E, Drury PL, et al. Can administrative health utilisation data provide an accurate diabetes prevalence estimate for a geographical region?. Diabetes Res Clin Pract, 139:59-71, 2018.
Chang HW, Lin YW, Lin MH, Lan YC, Wang RY. Associations between urate-lowering therapy and the risk of type 2 diabetes mellitus. PloS one, 14(1):e0210085, 2019.
Chen G, Khan N, Walker R, Quan H. Validating ICD coding algorithms for diabetes mellitus from administrative data. Diabetes Res Clin Pract, 89(2):189-195, 2010.
Chun, DS, Lund JL, Stürmer T. Pharmacoepidemiology and Drug Safety's special issue on validation studies. Pharmacoepidemiol Drug Saf, 28(2):123-125, 2019.
Chung CP, Rohan P, Krishnaswami S, McPheeters ML. A systematic review of validated methods for identifying patients with rheumatoid arthritis using administrative or claims data. Vaccine, 31:K41-K61, 2013.
Dart AB, Martens PJ, Sellers EA, Brownell MD, Rigatto C, Dean HJ. Validation of a pediatric diabetes case definition using administrative health data in Manitoba, Canada. Diabetes Care, 34(4):898-903, 2011.
Gavrielov-Yusim N, Friger M. Use of administrative medical databases in population-based research. J Epidemiol Community Health, 68(3):283-287, 2014.
Guo F, Moellering DR, Garvey WT. Use of HbA1c for diagnoses of diabetes and prediabetes: comparison with diagnoses based on fasting and 2-hr glucose values and effects of gender, race, and age. Metab Syndr Relat Disord, 12(5):258-268, 2014.
Guttmann A, Nakhla M, Henderson M, To T, Daneman D, Cauch Dudek K, et al. Validation of a health administrative data algorithm for assessing the epidemiology of diabetes in Canadian children. Pediatr Diabetes, 11(2):122-128, 2010.
Harris SB, Glazier RH, Tompkins JW, Wilton AS, Chevendra V, Stewart MA, et al. Investigating concordance in diabetes diagnosis between primary care charts (electronic medical records) and health administrative data: a retrospective cohort study. BMC Health Serv Res, 10(1):347, 2010.
Henriksen DP, Nielsen SL, Laurse CB, Hallas J, Lassen AT. How well do discharge diagnoses identify hospitalised patients with community-acquired infections?–a validation study. PLoS One, 9(3):e92891, 2014.
Herrett E, Thomas SL, Schoonen WM, Smeeth L, Hall AJ. Validation and validity of diagnoses in the General Practice Research Database: a systematic review. Br J Clin Pharmacol, 69(1):4-14, 2010.
Hsiao FY, Yang CL, Huang YT, Huang WF. Using Taiwan's national health insurance research databases for pharmacoepidemiology research. J Food Drug Anal, 15(2), 2007.
Hsing A, Ioannidis J. Nationwide population science: lessons from the Taiwan National Health Insurance Research Database. JAMA Intern Med, 175(9):1527‐1529, 2015.
Hux JE, Ivis F, Flintoft V, Bica A. Diabetes in Ontario: determination of prevalence and incidence using a validated administrative data algorithm. Diabetes care, 25(3):512-516, 2002.
Jiang J, Southern D, Beck CA, James M, Lu M, Quan H. Validity of Canadian discharge abstract data for hypertension and diabetes from 2002 to 2013. CMAJ Open, 4(4):E646, 2016.
Kee VR, Gilchrist B, Granner MA, Sarrazin NR, Carnahan RM. A systematic review of validated methods for identifying seizures, convulsions, or epilepsy using administrative and claims data. Pharmacoepidemiol Drug Saf, 21:183-193, 2012.
Koo BK, Lee CH, Yang BR, Hwang SS, Choi NK. The incidence and prevalence of diabetes mellitus and related atherosclerotic complications in Korea: a National Health Insurance Database Study. PloS one, 9(10):e110650, 2014.
Koram N, Delgado M, Stark JH, Setoguchi S,de Luise C. Validation studies of claims data in the Asia‐Pacific region: A comprehensive review. Pharmacoepidemiol Drug Saf, 28(2):156-170, 2019.
Lai CH, Chen YY. Type 2 diabetes mellitus and risk of hydrocephalus: a 5 year population-based follow-up study in Taiwan. J Diabetes Complicat, 30(3):426-431, 2016.
Lawrence JM, Black MH, Zhang JL, Slezak JM, Takhar HS, Koebnick C, et al. Validation of pediatric diabetes case identification approaches for diagnosed cases by using information in the electronic health records of a large integrated managed health care organization. Am J Epidemiol, 179(1):27-38, 2014.
Lee MC, Chiang CY, Lee CH, Ho CM, Chang CH, Wang JY, et al. Metformin use is associated with a low risk of tuberculosis among newly diagnosed diabetes mellitus patients with normal renal function: a nationwide cohort study with validated diagnostic criteria. PLoS One, 13(10):e0205807, 2018.
Leong A, Dasgupta K, Bernatsky S, Lacaille D, Avina-Zubieta A, Rahme E. Systematic review and meta-analysis of validation studies on a diabetes case definition from health administrative records. PloS one, 8(10):e75256, 2013.
Lin CC, Li CI, Hsiao CY, Liu CS, Yang SY, Lee CC, Li TC. Time trend analysis of the prevalence and incidence of diagnosed type 2 diabetes among adults in Taiwan from 2000 to 2007: a population-based study. BMC Public Health, 13(1):318, 2013.
Lin CC, Lai MS, Syu CY, Chang SC, Tseng FY. Accuracy of diabetes diagnosis in health insurance claims data in Taiwan. J Formos Med Assoc, 104(3):157-163, 2005.
Lipscombe LL, Hwee J, Webster L, Shah BR, Booth GL, Tu K. Identifying diabetes cases from administrative data: a population-based validation study. BMC Health Serv Res, 18(1):316, 2018.
Moores KG, Sathe NA. A systematic review of validated methods for identifying systemic lupus erythematosus (SLE) using administrative or claims data. Vaccine, 31:K62-K73, 2013.
Moreno-Iribas C, Sayon-Orea C, Delfrade J, Ardanaz E, Gorricho J, Burgui R, et al. Validity of type 2 diabetes diagnosis in a population-based electronic health record database. BMC Med. Inform Decis Mak, 17(1):34, 2017.
Pan HC, Sun CY, Wu IW, Tsai TL, Sun CC,Lee CC. Higher risk of malignant neoplasms in young adults with end‐stage renal disease receiving haemodialysis: A nationwide population‐based study. Nephrology, 24(11):1165-1171, 2019.
Quan H, Li B, Duncan Saunders L, Parsons GA, Nilsson CI, Alibhai A, et al. Assessing validity of ICD‐9‐CM and ICD‐10 administrative data in recording clinical conditions in a unique dually coded database. Health Serv Res, 43(4):1424-1441, 2008.
Rawson NS, D'Arcy C. Assessing the validity of diagnostic information in administrative health care utilization data: experience in Saskatchewan. Pharmacoepidemiol Drug Saf, 7(6):389-398, 1998.
Saczynski JS, Andrade SE, Harrold LR, Tjia J, Cutrona SL, Dodd KS, et al. A systematic review of validated methods for identifying heart failure using administrative data. Pharmacoepidemiol Drug Saf, 21:129-140, 2012.
Shrestha S, Dave AJ, Losina E, Katz J N. Diagnostic accuracy of administrative data algorithms in the diagnosis of osteoarthritis: a systematic review. BMC Med Inform Decis, 16(1):82, 2016.
Southern DA, Roberts B, Edwards A, Dean S, Norton P, Svenson LW, et al. Validity of administrative data claim-based methods for identifying individuals with diabetes at a population level. Can J Public Health, 101(1):61-64, 2010.
Teltsch DY, Fazeli Farsani S, Swain RS, Kaspers S, Huse S, Cristaldi C, et al. Development and validation of algorithms to identify newly diagnosed type 1 and type 2 diabetes in pediatric population using electronic medical records and claims data. Pharmacoepidemiol Drug Saf, 28(2):234-243, 2019.
Tsai CP, Lee JKW, Lee CTC. Type II diabetes mellitus and the incidence of amyotrophic lateral sclerosis. J Neurol, 266(9):2233-2243, 2019.
Tseng CH. Treatment with human insulin does not increase thyroid cancer risk in patients with type 2 diabetes. Eur J Clin Invest, 44(8):736-742, 2014.
Tseng CH. Metformin reduces gastric cancer risk in patients with type 2 diabetes mellitus. Aging (Albany NY), 8(8):1636, 2014.
Walkup JT, Townsend L, Crystal S, Olfson M. A systematic review of validated methods for identifying suicide or suicidal ideation using administrative or claims data. Pharmacoepidemiol Drug Saf, 21:174-182, 2012.
Wilson C, Susan L, Lynch A, Saria R, Peterson, D. Patients with diagnosed diabetes mellitus can be accurately identified in an Indian Health Service patient registration database. Public Health Rep, 116(1):45, 2001.
資料庫使用手冊Health90_糖尿病主題式資料庫。衛生福利資料統計處,衛生福利資料科學中心。
賴美淑。以健保申報醫療利用資料，建立台灣糖尿病人世代追蹤實証研究系統。行政院衛生署九十一年度委託研究計劃。2002。