| 摘要: | 近來患有睡眠呼吸障礙的人口與日俱增，而阻塞型睡眠呼吸中止症(obstructive sleep apnea, OSA)是常見且普遍之睡眠呼吸障礙疾病。其影響包含過度嗜睡、高血壓、心臟性猝死、糖尿病等。OSA確診需要經過整晚的多項睡眠生理檢查(Polysomnography, PSG)。但是完整的睡眠診斷費用昂貴、測試耗時且需專業人員進行，所需龐大的社會醫療資源也代表要對全部OSA疑似患者確診是極為困難的。目前有許多研究探索篩檢高風險OSA患者的方法，以期不必讓所有疑似患者皆進行PSG，以減少醫療資源之消耗。其中一個方法是建立臨床預測模型。多數臨床預測模型包括利用人體基礎參數變量(BMI、腰圍或頸圍等)或症狀（打鼾、呼吸暫停或喘氣等）作為預測模型變項，並將這些變項經由模式轉換為罹患OSA機率或分數，再藉以制訂機率或分數的切點，高於此切點即為高風險OSA患者。此種方法較其他篩檢方法的優勢包括只要一般健檢生理資料即可評估，有相對較低的篩檢成本，並可對大規模群眾進行快速篩檢。但以往只有少數研究曾比較多個不同臨床預測模型之敏感度與特異度，且大多數臨床預設模型研究只使用醫院病人做為資料來源，而尚未在其他特殊族群驗證其準確度。在職業駕駛或亞洲人等已知有較高OSA風險的族群上，如未比較多個模型的敏感度、特異度及未重新評估最佳切點，將使得臨床預測模型的應用受到限制。
本研究目的為以現有臨床預測模型，應用於職業族群上，比較利用原始建議切點所檢視之準確度差異，進而重新評估適用於職業族群之新切點，比較新舊切點之敏感度與特異度。另外本研究亦進行臨床預測模型在因子與係數的修正，以期增加模型敏感度與特異度。此外本研究亦計劃比較近年來所建立之二階段篩檢策略應用於職業族群的敏感度與特異度，希望可提供最適合應用於職場篩檢之臨床預測模型或篩檢策略。
本研究以特定職業族群為研究對象進行多項睡眠生理檢查，以確認職業勞工患有睡眠呼吸中止症的比率與嚴重程度。所有員工均以問卷及健康檢查方式收集篩檢模型所需數據，並以現有臨床預測模型計算罹患OSA機率或分數，再分別比較原始文獻建議切點及以Receiver operating characteristic(ROC)曲線建立之新切點。為進一步提升準確度，本研究進行臨床預測模型修正，修正方法分為只修正模型係數及修正模型納入因子和係數兩種。
本研究之結果發現原始建議切點應用於本研究族群時與原文獻準確度有落差。大部分模型結果為敏感度下降，特異度上升。而以ROC曲線建立之新最佳切點能使準確度提升。其預測輕度OSA最佳者為Rodsutti模型，而預測中重度OSA最佳者為Chai-Coetzer模型。無論修正模型係數及修正模型納入因子和係數皆可使模型準確度再提升。其中以修正係數方式較佳。修正模型中預測輕度OSA最佳者為Rodsutti修正係數模型，預測中重度OSA最佳者為MAP修正係數模型與Chai-Coetzer修正因子與係數模型。而加入血氧儀之二階段篩檢模型則可有較原臨床預測模型及修正臨床預測模型更為準確，其中準確度最佳者為Takegami二階段模型。
我們認為臨床預測模型運用至其他族群時需要重新評估最佳切點或修正模型，才可提升模型準確度。在測試的臨床預測模型中，Rodsutti修正係數模型、MAP修正係數模型與Chai-Coetzer修正因子與係數模型有較佳的結果。二階段模型則有較原模型及修正模型更佳準確度，而最佳者為Takegami二階段模型。未來期望能開發預測台灣職業駕駛AHI之預測模型，並將本研究已修正預測模型應用於其他族群驗證準確度。測試二階段篩檢策略之第二階段是否能以其他篩檢工具替代，期望更加提升模型之篩檢能力。
Obstructive sleep apnea(OSA) is a common sleep disorder giving risk to excessive daily sleepiness, hypertension, sudden cardiac death and diabetes. Polysomnography(PSG) is a golden tool in OSA diagnosed methods. However PSG is costly, taking time and requiring well trained personnel to operate in hospital. Suggesting applying PSG to the most of potential OSA patients is difficult because of the huge medical source and cost wasted. At the moment to reduce the cost, there are many studies searching for convenient methods to screen high risk OSA patient without PSG. One of the methods is clinical prediction model that transforms parameter such as basic human body parameter (ex: BMI, waist circumference and neck circumference) or symptoms (ex: snoring, apnea and gasping) into probability or score, and make cut points which filter out high risk OSA patients above points. The advantage of clinical prediction model is fast with large samples and less costly. But there are few studies to compare the sensitivity and specificity between models and only using hospital database without verifying by other specific groups. The application of clinical prediction model is limited without comparing the sensitivity and specificity and reassessing the cut point in different groups and accuracy decreased in known high OSA risk groups like drivers and Asians.
This study aimed to compare the accuracy of the original cut-point difference between the existing clinical prediction model was applied to occupational groups. Then re-evaluate the new clinical prediction model cut point used in occupational populations, and comparing sensitivity and specificity of the old and the new cut point. Also, to compare the sensitivity and specificity of the two stages screening strategy used in occupational groups, in order to provide the most suitable clinical prediction model or screening strategy of the workplace screening.
The focus of this study is to verify the existing clinical prediction model by worker groups who received PSG to confirm the OSA prevalence and severity. Using questionnaires and physical examination to collect parameters, then and to calculating clinical prediction model and comparing original cut points and new optimal cut points via receiver operating characteristic (ROC) curve. Clinical prediction model adjustment is further applied with two methods including model coefficient adjustment alone and both model parameters and coefficient adjustment for accuracy improving. The results reveal differences between our study and the original studies with decreasing sensitivity and increasing specificity. Unsurprisingly, new best cut points from ROC method increase the accuracy of all models. The best performing models to predict mild and moderate to severe OSA are Rodsutti model and Chai-Coetzer model, respectively. The model coefficient adjustment and model parameter and coefficient adjustment both slightly increase the prediction accuracy. The best performing prediction adjusted models for mild OSA and moderate to severe OSA are Rodsutti model, MAP model and Chai-Coetzer model, respectively. Two-stage strategy of screening OSA showed the better accuracy improving from original to adjusted model. Takegami two-stage model is the model with most accuracy.
In conclusion the clinical prediction models need cut point reassessment or model adjustment to increase the accuracy. Rodsutti coefficient adjustment model, MAP coefficient adjustment model, Chai-Coetzer coefficient and parameter adjustment model have better accuracy. Two-stage model provide better prediction accuracy than original and adjusted model and Takegami two-stage model has better result. In the future, developing a AHI prediction model of occupational workers. And to verify the accuracy of the adjusted prediction model by applying it to other groups are warranted. Moreover, whether using other screening tools such as actigraphy replacing the oximeter to increase the prediction accuracy in two-stage screening strategy also need to be investigated noteworthily. |
