臺灣博碩士論文加值系統

骨質疏鬆疾病確診的黃金標準工具為雙能量X光吸收測量儀(DXA)所測得骨質密度，當骨質密度T值低於-2.5SD則診斷為骨質疏鬆，但DXA在骨質疏鬆疾病的族群篩檢執行上並非一適合的工具。近年來，定量超音波工具在骨質密度篩檢上為一熱門工具，但討論其檢驗準確性的族群研究缺乏。
　　本研究蒐集了1999到2004年社區篩檢的骨質密度資料共41245筆，其中39512筆為非骨質疏鬆個案，1733筆為骨質疏鬆個案，得到定量超音波工具在族群資料的骨質密度T值小於等於-2SD時，有敏感度0.58(95%CI: 0.55-0.60)，特異度0.66 (95%CI: 0.64- 0.69)。考慮性別、年齡、BMI≧25、飲酒習慣、吸菸習慣、牛奶攝取頻率、咖啡攝取頻率、更年期狀態等骨質疏鬆的臨床危險因子對骨質密度的影響，帶入回歸模型以預測個人化的檢驗準確率，在此族群中有真陽性率預測範圍16.4%-91.6%，偽陽性率範圍4.7%-88.6%與陽性預測值範圍0.2%- 41.5%。在雙常態分布的假設下以ROC方法分析定量超音波工具在骨質密度範圍-8.0～+8.0間可能的骨質密度臨界值共23個切點的檢驗準確率，在不同族群分類下畫出ROC曲線，並得到AUC範圍0.59-0.71。遵從族群的ROC分析結果，考慮臨床因子年齡、性別及更年期狀態還有變項與骨質密度間的相關性，發展出個人化的檢驗結果預測值並建構出個人化的定量超音波工具的ROC曲線。利用模擬資料進行骨密度檢測工具ORAI對於骨質疏鬆檢測能力之評估，發現高年齡、體重輕者、無賀爾蒙使用者與停經女性亦為骨質疏鬆之危險因子，此模擬結果與本研究前述發現相符，並得到AUC：0.81。
　　本研究在骨質密度變動下建構出個人化的定量超音波工具ROC曲線作為思考點，達到提升定量超音波工具在檢驗骨質疏鬆疾病的敏感性，同時降低檢驗偽陽性率，並發展以定量超音波工具作為骨質疏鬆篩檢工具的個人化目標。

The gold standard for osteoporosis diagnosis is bone mass density (BMD) detected by dual energy X-ray absorptiometry (DXA), however DXA is not a suitable tool on population screening. Recently, the quantitative ultrasound (QUS) tool was developing as a BMD screening tool in population screening, but the accuracy of QUS tool has not yet been identified.
We collected population BMD screening data from 1999 to 2004, there are 41245 people with 39512 are non-osteoporosis and 1733 are diagnosed to be osteoporosis. Using QUS as BMD screening tool had sensitivity of 0.58(95%CI: 0.55-0.60) and sensitivity of 0.66 (95%CI: 0.64- 0.69) at cut-off point BMD T score≦-2SD.To predict individual accuracy of QUS tool we used regression model to analysis, dichotomous BMD value as dependent variable and risk factors based on FRAX which to be age、gender、BMI above or below 25、milk intake three days or more per week、coffee intake three days or more per week、before or after menopause as independent variables. ROC curve is to analysis the accuracy at possible cut-off point, and AUC range 0.59-0.71 at different population under binormal distribution. After modeling the effect of covariate on test result, we developed the predictive test result of individual and build individual ROC curve at each possible cut-off point of QUS tool.
Take individual information in to consideration would elevate the accuracy of QUS tool. Estimated the performance by regression model had AUC 0.81, this result performs similarly well as the simulation result which to be 0.79 in previous study. In this, we develop QUS tool to become a more accurate screening tool in BMD screening.

Chapter 1 評估超音波骨密度偵測準確性之必要性………………………… 1
Chpater 2. 文獻探討…………………………………………………………… 3
2.1檢驗準確度……………………………………………………………… 3
2.1.1偽陽性率與偽陰性率(True Positive Fraction and False Positive
Fraction) ………………………………………………………… 3
2.1.2 診斷概似函數比(Diagnostic Likelihood Ratio, DLR) ………… 5
2.1.3 估計Case-Control研究樣本…………………………………… 6
2.2準確度迴歸分析(Regression models for classification
probabilities) …………………………………………………………… 8
2.2.1真陽性率與偽陽性率之迴歸分析……………………………… 8
2.2.2陽性預測值之迴歸模型………………………………………… 9
2.3 接受者作業特徵曲線(Receiver Operating Characteristic Curve,
ROC curve) ……………………………………………………………… 11
2.3.1 非二分法( Non-binary)檢驗結果………………………………… 11
2.3.2二分法( Dichotomized) 診驗結果………………………………… 11
2.3.3 接受者作業特徵曲線( Receiver Operation Characteristic curve, ROC
curve) ……………………………………………………………… 12
2.3.3.1 ROC曲線之定義…………………………………………… 12
2.3.3.2 ROC曲線之數學性質……………………………………… 14
2.3.4 ROC curve的應用…………………………………………………… 17
2.3.5 曲線下面積 (Area under the ROC curve , AUC) ………………… 19
2.4 雙常態接受者作業特徵曲線(Binormal Receiver Operation Characteristic curve Binormal ROC curve) ………………………………………………… 22
2.4.1 雙常態接受者作業特徵曲線基本公式…………………………… 22
2.4.2 雙常態接受者作業特徵曲線下面積(The Binormal AUC)………. 24
2.4.3 雙常態分布假設…………………………………………………... 25
2.5 Modeling Covariate Effects on Test Results………………………… 27
2.6骨質疏鬆及骨質密度檢驗文獻回顧…………………………………… 29
2.6.1 流行病學………………………………………………………… 29
2.6.2 盛行率…………………………………………………………… 30
2.6.3 骨疏性骨折成因與好發處……………………………………… 30
2.7骨質密度檢測……………………………………………………………… 32
2.7.1 疾病診斷與T值計算…………………………………………… 32
2.7.2 臨床危險因子（Clinical Risk Factors, CRFs）………………… 34
2.8 骨質疏鬆臨床診斷工具………………………………………………… 36
2.8.1 雙能量X光吸收測量儀 (Dual-X-ray Ａbsorptiometry, DXA) ……36
2.8.2定量斷層掃描(Quantitative Computed Tomography, QCT) ………… 37
2.8.3定量超音波（Quantitative Ultrasound, QUS）…….………………… 38
2.8.4 臨床危險因子問卷工具 (Risk assessment questionnaires) ……… 38
2.9 Osteoporosis Prediction tool………………………………………………… 40
2.9.1 Predictive model for Osteoporosis…………………………………… 42
2.9.2 Predictive model for Osteoporosis Fracture……………………………44
2.10骨質疏鬆的種類及臨床治療方法成效評估……………………………… 46

Chapter 3. 材料與方法……………………………………………………………47
3.1 資料來源………………………………………………………………… 47
3.1.1 研究樣本……………………………………………………………… 47
3.1.2 骨質密度(BMD)檢測及陽性個案定義………………………………47
3.1.3變項蒐集……………………………………………………………… 48
3.2統計方法之應用………………………………………………………………50
3.2.1名詞定義……………………………………………………… 50
3.2.2 Delta method for deriving the variance of　estimates on likelihood
ratios…………………………………………………………… 51
3.2.3 logDLR+及logDLR－之共變異數(Covariance) ……………… 53
3.3 準確度迴歸分析(Regression models for classification probabilities) … 55
3.3.1 檢驗真陽性率與偽陽性率之迴歸模式…………………………… 55
3.3.2陽性預測值………………………………………………………… 57
3.4接受者作業特徵曲線( Receiver Operation Characteristic curve, ROC curve)…58
3.4.1接受者作業特徵曲線斜率與概似函數比…………………………… 58
3.4.2 Area under ROC curve (AUC) ……………………………………… 59
3.5雙常態接受者作業特徵曲線　(Binormal Receiver Operation Characteristic
curve, Binormal ROC curve) ……………………………………………… 62
3.5.1 雙常態接受者作業特徵曲線協率及概似函數比…………………. 63
3.5.2 雙常態分布接受者作業特徵曲線下面積 ( The binormal AUC ) … 65
3.6 Estimating the Roc curve…………………………………………………… 67
3.7 Modeling Covariate Effects on Test Results……………………………… 70

Chapter 4 Result ………………………………………………………………… 72
4.1 描述性結果……………………………………………………………… 72
4.2 疾病分類正確率(Disease-specific classification probabilities) ………… 73
4.2.1 超音波骨質密度偵測工具對照疾病狀態分析…………………… 73
4.2.2 定義檢驗結果分類錯誤機率(Misclassification probability) ……… 73
4.2.3 診斷概似函數比(Diagnostic likelihood ratios , DLR) ……………… 74
4.3 準確率之影響因子………………………………………………………… 76
　4.4陽性預測值之影響因子分析……………………………………………… 78
4.5雙常態分布接受者作業特徵曲線與曲線下面積………………………… 80
4.6 Empirical ROC curve…………………………………………………… 82
4.7 Modeling Covariate Effects on Test Results…………………………… 85

Chapter 5 Discussion……………………………………………………………. 88
5.1 主要方法及結果………………………………………………………… 88
5.1.1骨質疏鬆疾病臨床危險因子……………………………………… 88
5.1.2定量超音波及個人特質做為骨質疏鬆篩檢工具之意義，包括敏
感度提高及偽陽性個案減少……………………………………… 89
5.1.3 雙常態分布下最理想的(Optimization)偽陽性率與偽陰性率… 90
5.1.4 雙常態分布下ROC方法於骨質疏鬆檢測之應用……………… 91
5.1.5 研究限制………………………………………………………… 91
5.2 Conclusion………………………………………………………………… 93

參考文獻……………………………………………………………… 94

Armitage P Statistical Methods in Medical Research Forth edition. Blackwell Science Ltd Edited.

S M Cadarette， S B Jaglal， N Kreiger， W J McIsaac， G A Darlington， and J V Tu Development and validation of The Osteoporosis Risk Assessment Instrument To facilitate selection of women for bone densitometry CMAJ • MAY 2, 2000; 162 (9).

Tony Hsiu-Hsi Chen, D.D.S., Ph.D. Yueh-Hsia Chiu, M.Sc Dih-Ling Luh, Ph.D. Ming-Fang Yen, M.Sc. Hui-Min Wu, M.Sc. Li-Sheng Chen, M.Sc. Tao-Hsin Tung, M.Sc. Chih-Chung Huang, B.Sc. Chang-Chuan Chan, Sc.D. Ming-Neng Shiu, M.D., Ph.D. Yen-Po Yeh, M.D. Horng-Huei Liou, M.D., Ph.D. Chao-Sheng Liao, M.D., M.Sc. Hsin-Chih Lai, Ph.D. Chun-Pin Chiang, D.D.S., D.M.Sc. Hui-Ling Peng, M.D. Chuen-Den Tseng, M.D., Ph.D. Ming-Shyen Yen, M.D.Wei-Chih Hsu, M.D., M.Sc. Chih-Hung Chen, M.D. Community-Based Multiple Screening Model Design, Implementation, and Analysis of 42,387 ParticipantsTaiwan Community-Based Integrated Screening Group. 2004 American Cancer Society DOI 10.1002/cncr.20171 Published online 3 March 2004 in Wiley InterScience (www.interscience.wiley.com).

Compston JE, FRCPath FMedSci Clifford J Rosen MD FAST FACTS Osteoporosis Fourth edition

Cooper M, BMBCh PhD Our approach To osteoporosis screening and Treatment needs To change CMAJ.JUNE 17.2008.178(13)

Delong ER.,Delong DM., Clarke-Pearson DL. Comparing the Area Under Two or More Correlated Receiver Operating Characteristic Curves A Nonparametric Approach. Biometrics (1988) 44, 837-845.

Dodd LE., Pepe MS. Partial AUC Estimation and Regression. UW Biostatistics Working Paper Series University of Washington. Year 2003 Paper 181.

Donaldson MG., Palermo L., Schousboe JT.,Ensrud KE., Hochberg MC.,and Steven R. FRAX and Risk of Vertebral Fractures: The Fracture Intervention Trial. Cummings1 JOURNAL OF BONE AND MINERAL RESEARCH Volume 24, Number 11, 2009 Published online on May 4, 2009; doi: 10.1359/JBMR.090511_ 2009 American Society for Bone and Mineral Research

Hanley JA., Ph.D. Barbara |. McNeil, M.D., Ph.D.The Meaning and Use of the Area under a Receiver Operating Characteristic (ROC) Curve. RADIOLOY Vol .143 .N o. l .Pages29 36 .April. 1982.

Johnell OA., Kanis JA. An estimate of the worldwide prevalence, mortality and disability associated with hip fracture. Osteoporos Int (2004) 15: 897–902 DOI 10.1007/s00198-004-1627-0

Kanis JA., Borgstrom F., Laet CD.,Johansson H., Johnell O., Jonsson B., Oden A., Zethraeus N., Pfleger Nikolai B., Khaltae N. Assessment of fracture risk. Osteoporos Int (2005) 16: 581–589 DOI 10.1007/s00198-004-1780-5

Kanis JA., Johnell O., Oden A., Johansson H., McCloskey E. FRAX™ and the assessment of fracture probability in men and women from the UK. Osteoporos Int (2008) 19:385–397

Lim LS， MD， MPH， FACPM， Hoeksema LJ， MD，Sherin K， MD.Screening for Osteoporosis in The Adult U.S. Population ACPM Position Statement on Preventive Practice .MPH， FACPM， and The ACPM Prevention Practice Committee Am J Prev Med 2009;36(4)

Pepe MS，Professor of Biostatistics, Universuty of Washington and Fred Hutchinson Cancer Research Center. The Statistical Evaluation of Medical Tests for Classification and Prediction. OXFORD University Press,2004

Sandhu SK., Nguyen ND., Center JR., Pocock NA., Eisman JA., Nguyen TV. Prognosis of fracture: evaluation of predictive accuracy of the FRAX™ algorithm and Garvan nomogram. Osteoporos Int (2010) 21:863–871 DOI 10.1007/s00198-009-1026-7.

SINGER A, MD. Osteoporosis Diagnosis and Screening. Clinical Cornerstone _ MANAGEMENT OF OSTEOPOROSIS _ Vol. 8, No. 1

Tzay SY., Chen YR., Chen YJ., Chang CY., Ng HT .Osteoporosis: prevalence in Taiwanese women. Osteoporosis Int (2004) 15: 345–347 DOI 10.1007/s00198-003- 1509-x,