published_or_final_version / abstract / toc / Diagnostic Radiology / Master / Master of Research in Medicine

目的：本研究主要目的是建立預測模型來判定矽肺發生的累積風險從而推薦適宜的胸片照射年限並從而評估常規監測和推薦監測策略的成本效益。此外，本研究還評價了常規診斷和驗證性診斷的符合度以及在驗證性診斷中邀請的三位專家之間的符合度。 / 方法：總計有3492男性接塵工人在1964年1月1日到1974年12月31日期間進入本隊列並隨訪至2008年12月31日。不同閱片專家根據中國最新塵肺病診斷標準 （GBZ70-2009））分別閱片總計9084張。對專家之間閱片結果的兩兩比較和兩種診斷結果的比較均采用Cohen’s Kappa檢驗。應用三種篩選方法（強制所有變量同時進入模型，後退逐步篩選，以及Least Absolute Shrinkage and Selection Operator (LASSO)篩選。LASSO模型作為最優模型，以分數量表的方式來表達。根據分數，把工人分成高、中、低危險組，並估計這三組不同危險水平工人的累積危險度。運用0.1% 累積危險度來判定不同危险的工人及不同期別的矽肺病人的射線照射年限。多狀態Markov模型用於計算矽塵暴露工人不同狀態的年轉移概率，並應用Markov成本效益分析方法來估計每獲得一個生命年的成本效益。 / 結果：截至2008年底，本矽塵暴露隊列共計發現298例矽肺病人（累計發病率為8.53%），死亡1347例（死亡比例為38.57%）。本研究發現常規診斷和驗證性診斷有很好的符合度 （Kappa值為0.89， 95%可信區間為0.88-0.91）。基於LASSO模型的分數量表具有很好的診斷識別能力 （ROC曲線下面積為0.83， 95%可信區間為0.81-0.86）。根據0.1%累積危險度標準，我們判定低危險組工人第一次射線照射的時間為第11年，推薦每兩年隨訪一次；中等危險組工人和高危險組工人的第一次射線照射時間分別為第11年和第5年，推薦每年隨訪一次。矽肺病人未晉級到三期以前均一年隨訪一次。矽塵暴露工人的年轉移概率為：從健康狀態向疑似病例轉移的概率為0.0198，從疑似病例向一期矽肺轉移的概率為0.038，從一期矽肺向二期矽肺轉移的概率為0.0516，從二期矽肺向三期矽肺轉移的概率為0.059，從三期矽肺向死亡轉移的概率為0.18。在1964到2008年間，診斷一例矽肺病例平均花費醫療成本為21853.11美元，非醫療成本為5993.30美元。模擬10,000矽塵暴露工人在未來40年按照當前的狀態轉移概率，應用常規的職業健康檢查為手段獲得一個生命年的成本效益為43.60美元，應用推薦的職業健康檢查為手段獲得一個生命年的成本效益為46.99美元。 / 結論：本研究在最優預測模型的基礎上為不同矽肺危險度的矽塵暴露工人首次提供了科學的證據來判定射線照射的適宜年限，亦為未來矽塵暴露工人的職業健康監測提供了科學理論依據，雖然本研究推薦的監測策略獲得同常規策略相類似的成本和效益。 / Objectives: The primary objective was to develop prediction models for determining the optimal intervals of chest radiographic surveillance for workers exposed to silica dust; the second primary objective is to assess the cost per case identification and compare the cost per life year gained under routine medical surveillance program with that under the recommended program for workers exposed to silica dust in China. In addition, the inter-rater agreement amongst three invited radiologists on rereading the chest radiographs and the agreement between the original diagnoses of silicosis (from routine reports) and the verified diagnoses reassessed by the three experts were also evaluated. / Methods: A total of 3492 male workers exposed to silica dust in an iron ore during the period 1964 - 1974 were recruited into this retrospective cohort study. All cohort members were followed up through the end of 2008 to observe the occurrence of silicosis and overall profile of mortality. All 9084 chest X-ray films were reread by three radiologists who had been qualified as experts at the national level according to the Chinese National Diagnostic Criteria of Pneumoconiosis (GBZ70-2009). The diagnosis of silicosis made by the panel of these three invited experts was referred to the “verified diagnosis“. Cohen’s Kappa test was used to test inter-rater agreements of three invited readers on chest radiographs and the agreement on the diagnosis of silicosis obtained from routine medical surveillance (i.e., the original routine diagnosis) was compared with those verified by the 3 qualified readers (i.e., the verified diagnosis). The multivariate Cox’s proportional hazard regression models were developed to predict the silicosis occurrence based on three selection approaches entry of all predictors at the same time, backward stepwise selection, and Least Absolute Shrinkage and Selection Operator (LASSO) selection. The LASSO model showed the best model fit which was thus regarded as the final model for predicting a score chart. / Based on this practically used score chart, we then classified workers into three groups of different risk levels of silicosis (low, moderate, and high). We estimated the cumulative risk of silicosis over years of follow-up for these three groups of workers at different risk levels. We used 1 per thousand of cumulative risk for developing silicosis as a “benchmark“ to determine the intervals of radiologic surveillance for workers with different risks of silicosis. Multi-state Markov model was used to calculate the transition probabilities of different states of silicosis and the analysis on cost and effectiveness was performed. / Results: By the end of 2008, the cumulative incidence rate of silicosis was 8.53% (298 silicosis cases) and a total of 1347 deaths (38.57%) were observed. / Good inter-rater agreements were observed amongst three invited radiologists for rereading all the chest films. Kappa value for the agreement between the original diagnoses and the verified diagnoses was 0.89 (95% confidence interval [95%CI], 0.88-0.91). / The model with the best fit was LASSO Cox model which showed a good discrimination with an area of 0.83 (95%CI, 0.81-0.86) under the receiver operating characteristic (ROC) curve. We classified workers into 3 risk groups according to the score chart obtained from the LASSO Cox model, and found the observed probabilities matched well to the predictions. According to 1 per thousand “benchmark“, we can determine that the initial interval of radiographic surveillance for workers in the low risk group was 11 years and a subsequent biyearly examination was recommended. The initial examination interval was 11 years and 5 years respectively for workers in the middle and high risk group, and then a yearly examination was recommended. For patients with silicosis, an annual radiological surveillance program was recommended regardless of the stage of pneumoconiosis. / According to results from multi-state model, we estimated that the yearly transition probability was 0.0198 for silica dust exposed workers from healthy state to the suspected silicosis cases (sojourn time = 47 years), 0.0338 from suspected silicosis cases to silicosis stage one (sojourn time = 23 years), 0.0516 from silicosis stage one to stage two (sojourn time = 9 years), 0.059 from silicosis stage two to stage three (sojourn time = 6 years), and 0.18 from silicosis stage three to death (sojourn time = 5 years). / During the period 1964 to 2008, the average direct medical cost spent on identifying one silicosis case was US$ 21853.11 and the non-medical cost for identifying one case was US$ 5993.30 per case. The estimated medical cost regarding per life year gained was US$ 43.60 under the routine medical surveillance program and it would be US$ 46.99 if the newly recommended surveillance program is adopted. / Conclusion: This study is the first to provide scientific evidence on determining the optimal intervals of radiographic surveillance for workers at different risk levels of silicosis based on the ‘best’ prediction model. Although our study revealed similar cost and effectiveness for using the recommended occupational health examination strategy compared with the routine program, this study is the first to provide scientific theory for guiding evidence-based occupational medical surveillance on workers exposed to silica dust in the world. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Chen, Minghui. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2012. / Includes bibliographical references (leaves 195-210). / Abstract also in Chinese. / Abstract (English) --- p.i / Abstract (Chinese) --- p.v / Acknowledgements --- p.vii / List of contents --- p.ix / List of tables --- p.xv / List of figures --- p.xviii / List of main abbreviations --- p.xx / Chapter Section I --- Introduction and Literature Review --- p.1 / Chapter Chapter 1 --- Introduction --- p.2 / Chapter Chapter 2 --- Literature Review of Medical Examination, Prediction model and Economic Evaluation in Silicosis --- p.7 / Chapter 2.1 --- The aims of this literature review --- p.7 / Chapter 2.2 --- Search strategies and selection criteria --- p.7 / Chapter 2.3 --- Searching results --- p.8 / Chapter 2.4 --- Critical appraisal criteria and quality of selected studies --- p.9 / Chapter 2.4.1 --- Critical appraisal criteria --- p.9 / Chapter 2.4.2 --- Quality of selected studies --- p.10 / Chapter 2.5 --- Overview of effectiveness of chest radiography in medical surveillance of silicosis for workers exposed to silica dust --- p.15 / Chapter 2.5.1 --- Occupational medical surveillance for workers exposed to silica dust --- p.15 / Chapter 2.5.2 --- Comparison of CT or HRCT and chest radiography --- p.16 / Chapter 2.5.3 --- Comparison of digital radiography (DR) and chest radiography --- p.17 / Chapter 2.5.4 --- Other tests to be relevant to silicosis diagnosis --- p.23 / Chapter 2.5.5 --- The effectiveness of chest radiography in medical surveillance and diagnosis of silicosis --- p.24 / Chapter 2.5.6 --- Comparison between the ILO Classification and the Chinese Diagnostic criteria of pneumoconiosis --- p.25 / Chapter 2.6 --- Overview of application of prediction model in silicosis and a review on methodology in prediction model --- p.32 / Chapter 2.6.1 --- Application of prediction model in occupational diseases --- p.32 / Chapter 2.6.2 --- Overview of application of predicting model in pneumoconiosis including silicosis in China in recent 10 years --- p.34 / Chapter 2.6.3 --- Development of prediction model and the applications from practical perspectives --- p.35 / Chapter 2.7 --- A review on economic evaluation in occupational diseases and the screening interval analyses --- p.42 / Chapter 2.7.1 --- An overview on economic evaluation in pneumoconiosis --- p.42 / Chapter 2.7.2 --- Overview of economic evaluation in occupational health and safety and screening interval analyses --- p.44 / Chapter 2.7.3 --- Overview for methodology of performing CEA --- p.45 / Chapter 2.8 --- Research gaps were found from this literature review --- p.52 / Chapter Section II --- Objectives and Methods --- p.53 / Chapter Chapter 3 --- General aims and objectives --- p.54 / Chapter 3.1 --- General aims --- p.54 / Chapter 3.2 --- Primary objectives --- p.54 / Chapter 3.3 --- Secondary objective --- p.54 / Chapter Chapter 4 --- Methodology and Research Plans --- p.55 / Chapter 4.1 --- Study Design --- p.55 / Chapter 4.2 --- The cohort --- p.55 / Chapter 4.3 --- Follow-up --- p.58 / Chapter 4.4 --- Data Collection --- p.58 / Chapter 4.4.1 --- Baseline information --- p.58 / Chapter 4.4.2 --- Diagnosis of silicosis and the verification --- p.59 / Chapter 4.4.3 --- Occupational hygiene monitoring data --- p.60 / Chapter 4.4.4 --- Cost data of medical examination --- p.61 / Chapter 4.5 --- Data Entry and Data Analyses --- p.62 / Chapter Section III --- Results and Discussions --- p.65 / Chapter Chapter 5 --- Description of the cohort --- p.66 / Chapter 5.1 --- Cohort recruitment --- p.66 / Chapter 5.2 --- Baseline characteristics --- p.69 / Chapter 5.3 --- Change of respirable silica dust concentration over time --- p.71 / Chapter 5.5 --- Occurrence of silicosis --- p.73 / Chapter 5.5.1 --- Basic characteristics of silicosis patients --- p.73 / Chapter 5.5.2 --- Trend of silicosis occurrence with calendar year --- p.78 / Chapter 5.5.3 --- Trend of silicosis occurrence with age of entering the cohort --- p.78 / Chapter 5.5.4 --- Trend of silicosis occurrence with cumulative exposure to respirable silica dust --- p.78 / Chapter 5.6 --- Survival distribution at different respirable silica dust exposure levels --- p.79 / Chapter 5.7 --- A summary of the results in Chapter 5 --- p.82 / Chapter Chapter 6 --- Agreement between the routine diagnosis of silicosis and the verified ‘new panel’ diagnosis --- p.83 / Chapter [Summary] --- p.83 / Chapter 6.1 --- Background --- p.85 / Chapter 6.2 --- Methodology --- p.86 / Chapter 6.2.1 --- The routine and the verified diagnosis of silicosis --- p.86 / Chapter 6.2.2 --- Inter-rater agreement --- p.87 / Chapter 6.3 --- Results --- p.89 / Chapter 6.3.1 --- Technical quality of chest X-ray films --- p.89 / Chapter 6.3.2 --- Inter-rater agreement amongst readers --- p.89 / Chapter 6.3.3 --- Agreement between the routine and the verified diagnosis of silicosis --- p.93 / Chapter 6.3.4 --- Agreement of the progression of silicosis between the routine and verified diagnosis --- p.95 / Chapter 6.4 --- Discussion --- p.97 / Chapter Chapter 7 --- Developing prediction model for determining the optimal intervals of chest radiographic examinations for workers at different risks of silicosis --- p.100 / Chapter [Summary] --- p.100 / Chapter 7.1 --- Background --- p.102 / Chapter 7.2 --- Methods --- p.104 / Chapter 7.2.1 --- The cohort and outcome determination --- p.104 / Chapter 7.2.2 --- Developing prediction models for silicosis --- p.107 / Chapter 7.2.3 --- Coding of Predictors --- p.113 / Chapter 7.3 --- Results --- p.118 / Chapter 7.3.1 --- Model Specifications --- p.118 / Chapter 7.3.2 --- Stepwise Selection and LASSO selection --- p.119 / Chapter 7.3.3 --- Model Validations: Stability and Optimism --- p.119 / Chapter 7.3.4 --- Model Presentations --- p.126 / Chapter 7.3.5 --- Cut-off point of follow up year for determining examination intervals --- p.130 / Chapter 7.4 --- Discussions --- p.136 / Chapter Chapter 8 --- Transition probabilities of multi-states for workers with silica dust exposure --- p.141 / Chapter [Summary] --- p.141 / Chapter 8.1 --- Background --- p.143 / Chapter 8.2 --- Methodology of multi-state model --- p.145 / Chapter 8.2.1 --- Survival data and multi-state model --- p.145 / Chapter 8.2.2 --- Markov model and transition states --- p.151 / Chapter 8.2.3 --- Model assessment --- p.153 / Chapter 8.3 --- Results --- p.154 / Chapter 8.3.1 --- Initial values specification and estimates of intensity matrix --- p.154 / Chapter 8.3.2 --- Transition probability matrix, mean sojourn times, and survival situation --- p.159 / Chapter 8.3.3 --- Observed and expected prevalence of each state for Model assessment --- p.163 / Chapter 8.4 --- Discussion --- p.165 / Chapter Chapter 9 --- Cost effectiveness analysis of occupational medical surveillance for workers exposed to silica dust --- p.168 / Chapter [Summary] --- p.168 / Chapter 9.1 --- Background --- p.170 / Chapter 9.2 --- Methodologies --- p.171 / Chapter 9.2.1 --- Costs and effectiveness --- p.171 / Chapter 9.2.2 --- Cost per silicosis identification estimation in the iron ore during 1964 to 2008 --- p.172 / Chapter 9.2.3 --- Cost effectiveness analysis in the Markov model --- p.173 / Chapter 9.3 --- Results --- p.176 / Chapter 9.3.1 --- Cost estimation and cost per silicosis identification in the iron ore cohort --- p.176 / Chapter 9.3.2 --- Cost effectiveness analysis in the Markov model --- p.181 / Chapter 9.4 --- Discussion --- p.187 / Chapter Section IV --- Conclusions and Implications --- p.191 / Chapter Chapter 10 --- Conclusions, implications, and recommendations --- p.192 / Chapter 10.1 --- Conclusions --- p.192 / Chapter 10.2 --- Implications and recommendations --- p.193 / Reference list --- p.195 / Chapter Appendix I --- Chest Radiographic Imaging of Different Diagnostic Criteria for Pneumoconiosis in China --- p.211 / Chapter Appendix II --- Diagnosis Stages among Different Diagnostic Criteria for Pneumoconiosis in China --- p.212 / Chapter Appendix III --- Publications in journals and international conferences during the PhD study --- p.213 / Chapter Supplement I --- Syntax for test proportionality of Cox model in R survival package and LASSO model in R penalized package --- p.215 / Chapter Supplement II --- Guideline of applying the prediction model in practice --- p.216 / Chapter Supplement III --- Syntax for multi-state model in R msm package --- p.221 / Chapter Supplement IV --- An example for cost estimation of adjusting inflation and exchanging --- p.222 / Chapter Supplement V --- Cost estimation of workers, suspected silicosis cases and silicosis patients in the iron ore during 1964 - 2008 --- p.223 / Chapter Supplement V (Continued) --- Cost estimation of workers, suspected silicosis cases and silicosis patients in the iron ore during 1964 - 2008 --- p.224 / Chapter Supplement VI --- Number of deaths for all cause of death in the iron ore cohort until 2008 --- p.225 / Chapter Supplement VII --- Decision tree of Markov model in the study --- p.226 / Chapter Supplement VII (Continued) --- Decision tree of Markov model in the study --- p.227