失智症早期篩檢偵測工具效度及成本效益之統計評估

Statistical Evaluation of Validity and Cost-effectiveness of Two Screening Tools for Early Detection of Dementia

盧韻如(Yun-Ru Lu)

指導教授 ： 劉宏輝

共同指導教授 ： 陳秀熙(Hsiu-Hsi Chen)

國立臺灣大學/公共衛生學院/流行病學與預防醫學研究所/博士(2015年)

https://doi.org/10.6342/NTU.2015.11233

摘要

研究背景
隨著老年人口逐漸增加同時，失智症也有同時增加的趨勢，因此考慮一個符合有效益且有效率的社區型失智症篩檢是非常重要的。在此之前，我們需要一系列的研究，包括流行病學關於失智症早期偵測的認知程度，社區失智症篩檢可行工具的效度，和以社區為基礎的失智症篩檢計畫的經濟評估，及發展出一系列以理論為基礎的統計方法。
研究目的
本論文研究目的有三：(1) 利用一新的估計盛行率/發生率比率為基礎的統計回歸模型，來測量早期偵測，對於失智症發生的平均時間的影響(2)利用一貝氏臨床推理模型，來評估同時合併認知功能檢測(MMSE)及訊息提供問卷(AD8)二種篩檢工具，是否會加社區失智症篩檢診斷的精確度 (3)利用馬可夫決策分析模型，評估社區失智症篩檢中，其不同篩檢模的成本效益分析
研究材料及方法
利用2000-2003年期間，全民健保費用申報的大型資料庫，我們分析出65歲以上有失智症病人，且符合ICD-9-CM 編碼為 290, 290.40, 331.0者，共99,609位。另一資料庫來自2013年社區主動型篩檢，65歲以上參與者共183位。我們收集關於年紀、性別、及就醫地點等相關訊息。利用貝氏方法估出盛行率與發生率比率，來反映一般民眾對於失智症的認知程度。使用貝氏以盛行率/發生率比率為基礎的統計回歸模型，來做盛行率與發生率比率的校正，用此反映對失智症的認知程度。第二部分，在 AD8及MMSE的效度評估研究中，我們針對2013年台南社區整合型篩檢計畫，50歲以上民眾，評估AD8及MMSE二種篩檢工具其單獨使用、或平行檢定、或序列檢定，其效度使否有差異? 利用受試者工作特徵曲線（ROC）來探討 AD8及 MMSE在失智症及輕度認知功能缺損的預測功能。在缺乏黃金診斷標準下，利用一貝氏臨床推理模型來估計不同失智症篩檢工具模式的效度。第三部分，架構一馬可夫決策分析模型，來模擬有做失智症社區篩檢的族群其 10年成本及效益，並比較不同篩檢工具，包括 AD8、MMSE、平行檢定、序列檢定，與無篩檢的族群的成本及效益分析。我們使用一五階段馬可夫模型來模擬失智症進展，利用已發表的國內外文獻所提供參數(轉移機率、各階段所需醫療費用及照顧費用)。效益部分以品質調整人年命 (QALY)來測量，並含3%每年折扣率，並估出增加成本效用比 (ICUR)。利用500次蒙特卡羅模擬，得到從整體社會觀點畫出的分散成本效益面、及接受曲線，此指標可以顯示在不同付費意願閥值下，可以產生符合成本效益的百分比。
結果
失智症盛行率、發生率在健保資料庫(被動性篩檢組)估計值分別為2.91%及1.83%。利用盛行率/發生率比率為基礎的統計回歸模型估計校正區域後，得到， 被動性篩檢中，65歲以上老人盛行率對發生率的比率，男性從在65-79歲族群 1.20(1.15-1,24)到90歲以上族群3,27 (3.13-3.41)。盛行率/發生率比率發現，65歲以上老人盛行率對發生率的比率在地理區域北區最高，東區最低。被動性篩檢組，校正年紀、性別、區域對失智症影響後，發現盛行率對發生率的比率65-79歲族群1.45(1.43-1.47) 到80歲以上族群1.64 (1.61-1.66)。參與社區失智症篩檢組 (主動性篩檢組)，發現其盛行率對發生率的比率65-79歲族群4.23(2.68-6.69) 到80歲以上族群4.77 (3.02-7.54)。
AD8單獨使用對失智症敏感度及特異度分別為64.71% 及87.89%，MMSE單獨使用在校正教育程度的情況，對失智症敏感度及特異度分別為41.18% 及84.50% 。兩種篩檢工具合併使用，平行檢定的敏感度及特異度分別為88.89% 及70.16% ，序列檢定的敏感度及特異度分別為50% 及93.02%。AD8單獨使用對失智症+輕度認知功能障礙，敏感度及特異度分別為25.74% 及90.70%。所有篩檢模式，對於輕度認知功能障礙的敏感度，除了切點切在26分以外 ，其他組模式皆顯示敏感度較差，特異度尚可。利用受試者工作特曲線（ROC）來探討 AD8及 MMSE在失智症及輕度認知功能缺損的預測功能。結果發現合併 AD8和MMSE (平行檢定組)的 曲線下面積(AUC)為 82.3% (75.1%-89.4%)產生比AD8 (AUC=73.3% (60.7%-85.9%))單獨使用或 MMSE (AUC=77.4 %( 67.6%-87.3%)單獨使用或系列檢定(AUC= 67.6% (53.4%-81.8%))更好的預測功能。利用貝氏分析方法，可得到平行檢定提高敏感度(97.2%)相較於 MMSE (82.2%)或 AD8單獨使用(84.1%)：序列檢定提高特異度度(96.8 %) 相較於 MMSE (86.1%)或 AD8 (77.1%)單獨使用。關於經濟評估結果如下：只考慮篩檢及醫療支出的直接成本，則成本效用比 (ICUR)在AD8、MMSE、平行檢定、序列檢定中分別為每一人年為美金401.4、457.7、409.8、499.2元。同時考慮間接成本，在分散成本效益面評估為大約80％的模擬值在第四象限（顯性）。由四種篩檢模式中可知，如果政府願意付費閥值到達美金20000元，則可以得到88-94%成本效益。如果，只考直接成本，評估為大約40％的模擬值在第四象限（顯性），當願意付費閥值到達美金20000元，則可以得到93-99%成本效益。
結論
從預防由失智症造成失能和死亡的臨床方面來看，我們的研究，藉由根據所估計盛行率/發生率比率在社區篩檢模式(主動性偵測) 相較於健保照護體系(被動性偵測) 較高，已證實在一般健保照護體系，失智症的認知程度偏低。合併使用AD8和 MMSE在社區失智症篩檢，可提高工具敏感度。最後，使用AD8和 MMSE在社區失智症篩檢 是符合成本效益的：相較於無篩檢組幾乎接近省錢的。最符合成本效益的是合併使用AD8和 MMSE的平行檢定模式。我們的研究結果可應用於健康照護政策評估，及其他有興趣於發展社區失智症篩檢的計畫，進而減少失智症照護的支出。從方法學的角度，我們發展出三種創新方法，包括(1) 估計盛行率/發生率比率為基礎的統計回歸模型。(2) 利用貝氏臨床推理模型來估計不同失智症篩檢工具模式的效度。 (3) 馬可夫決策分析模型，評估社區失智症篩檢中結合AD8和 MMSE 的成本效益分析。

關鍵字

失智症 ； 篩檢 ； 敏感度 ； 特異度 ； 成本效益

並列摘要

Background
As there is an increasing trend in the morbidity of dementia when aging population has been increasing, considering an effective and efficient community based dementia screening programs is of paramount important. Before doing so a series of studies would be required to embrace various aspects including epidemiological assessment related to awareness of early detection of dementia, the validity of feasible screening tool for community-based screening for dementia, and economic evaluation of community-based screening program with the development of a series of theoretically-sound statistical methods.
Aims
This thesis aimed to (1) quantify the impact of early detection related to awareness on the average duration of disease based on the measurement of the ratio of prevalence to incidence of dementia with a newly proposed P/I-ratio-based statistical regression model; (2) assess the validity of the accuracy of the early detection of dementia with cognitive test (MMSE) and informant questionnaire (AD8) alone and particularly in the combination applied to a community-based dementia screening with Bayesian clinical reasoning model; and (3) perform cost-effectiveness of community-based dementia screening program with various screening strategies proposed in the second aim with the Markov decision tree model.
Materials and Methods
By using a large-scale, claimed data of the National Health Insurance (NHI) database between 2000 and 2003 in Taiwan, we identified 99,609 patients age over 65 years with dementia (ICD-9-CM code 290, 290.40, and 331.0). The other data source included a total of 183 subjects aged over 65 years participating in an active dementia survey conducted in 2013. Information on age, gender, and geographic areas were also collected. Bayesian P/I-ratio-based statistical regression method was used to estimate the adjusted prevalence/incidence (P/I) ratios of dementia to reflect the awareness of dementia. For the validity of AD8 and MMSE as well as the combination of the two tools in the parallel and the serial mode, we applied the two screening tools simultaneously in a community-based screening program for dementia to 282 Tainan residents aged over 50 years in 2013. Receiver operating characteristic curves (ROC) were applied to explore the performance of different screen modalities for prediction of MCI or dementia. Bayesian clinical reasoning method was used to estimate the performance of screening modalities in the absence of golden-standard diagnosis.
The Markov decision analysis was conducted to investigate the cost-utility of community-based screening of dementia over a 10-year period to compare different screening tools (AD8, MMSE, parallel and serial test of the two) with no screening. We used a five-state Markov model to simulate the progression of dementia. Disease transition probabilities and costs of different stages were extracted from literatures. The main outcome measure was cost per quality-adjusted life-year gained with a 3% annual discount rate. The scattered cost-effectiveness plane (CE plane) and acceptability curve are presented given a 500 Monte Carlo simulated samples.
Results
The prevalence and incidence rate of dementia based on passive survey were estimated as 2.91% and 1.83 %, respectively. The results with the application of Bayesian P/I-ratio-based statistical regression model show the adjusted P/I ratio increased from 1.20 (1.15-1.24) for 70-74 age group to 3.27 (3.13-3.41) for 90+ age group in males. The P/I ratio was the highest in northern area the lowest in eastern area. After controlling for age, gender, and geographic area, the adjusted P/I ratio increased from 1.45 (1.43-1.47) for 65-79 age group to1.64 (1.61-1.66) for 80+ age group through passive detection method (health insurance system). The corresponding figures increased from 4.23 (2.68-6.69) for 65-79 age group to 4.77 (3.02-7.54) for 80+ age group in active community-based survey. The sensitivity and specificity of the sole use of AD8 in dementia screening were 64.71% and 87.89%. The sensitivity and specificity of the sole use of MMSE in dementia with adjustment for education level were 41.18% and 84.50%. The combination of AD8 with MMSE in parallel mode yielded 88.89% of sensitivity and 70.16% of specificity. The combination of AD8 with MMSE in serial mode yielded 50.00% of sensitivity and 93.02% of specificity. The estimates of sensitivity and specificity of using AD8 test alone for MCI plus dementia were 25.74% and 90.70%. All the estimates of sensitivity for all the modes except the MMSE with 26 of cutoff for detecting MCI were poor and the specificity was moderate. By combining prior information derived from the results of previous studies with Bayesian approach, the results show the parallel mode had higher sensitivity (97.2%) than either MMSE (82.2%) or AD8 (84.1%) alone . Besides, the serial test had higher specificity (96.8 %) than AD8 (77.1%) or MMSE (86.1%) alone. ROC curve showed that the combination of MMSE and AD8 in the parallel mode (AUC=82.3% (75.1%-89.4%)) produced a more accurate prediction of dementia than the use of AD8 (AUC=73.3% (60.7%-85.9%)) and MMSE (AUC=77.4 %( 67.6%-87.3%) alone and also the serial mode (AUC= 67.6% (53.4%-81.8%)). Regarding economic evaluation, if only direct cost on screening and medical expenditure were considered, the ICURs for AD8, MMSE, parallel test, and sequential test were $401.4, $457.7, $409.8, and $499.2 per QALY gained, respectively. The scatted CE plane suggested that around 80% simulated sit in fourth quadrant (dominant) when indirect cost was considered. The probability of being cost-effective was 88-94% given willingness-to-pay (WTP) at $20,000 for the four screening scenario. The corresponding figures for being dominant and cost-effective at WTP at $20,000 when only direct cost taking into account were 40% and 93-99%, respectively.
Conclusions
From the practical aspect of prevention of disability and death from dementia, low awareness of dementia has been ascertained in routine health insurance health care system as the P/I ratio of community-based survey (active detection method) was greater than that of health insurance heath care system (passive detection method). The combination tests of MMSE and AD8 could improve diagnostic accuracy in the community dementia screening. Community-based screening for dementia with AD8 and MMSE is more cost-effective and almost near cost-saving compared with no screening program. The most economic screening strategy is the parallel mode of combining AD8 with MMSE in comparison with other modes.
From the methodological viewpoint, there are three novelties of the methodological development here, including a P/I-ratio-based regression model, the application of Bayesian model for multiple detection modalities, and the development Markov cycle decision tree model for economic evaluation of population-based screening program with AD8 in combination with MMSE. The empirical data together with the development of theoretically-sound statistical method provides a new insight into how to conduct an effective and efficient community-based screening for dementia.

並列關鍵字

dementia ； screening ； sensitivity ； specificity ； cost-effectiveness

參考文獻

1. Wolfson C, Wolfson DB, Asgharian M, et al. A reevaluation of the duration of survival after the onset of dementia. New England Journal of Medicine 2001;344:1111-1116.

2. Neumann P, Araki S, Arcelus A, et al. Measuring Alzheimer’s disease progression with transition probabilities Estimates from CERAD. Neurology 2001;57:957-964.

3. American Psychiatric Association A, Association AP. Diagnostic and statistical manual of mental disorders. 1994.

4. Larson EB, Kukull WA, Katzman RL. Cognitive impairment: dementia and Alzheimer's disease. Annual review of public health 1992;13:431-449.

5. Breteler MM, Claus JJ, van Duijn CM, Launer LJ, Hofman A. Epidemiology of Alzheimer's disease. Epidemiologic Reviews 1992;14:59-82.