精神分裂症的治療療程至少需持續4-8週，因為疾病的複雜性與病人對藥物反應的差異性，醫師很難在治療前依據病人臨床症狀與個人特質來評估病人對藥物反應的結果，為了避免病人接受無療效的藥物治療，建構一個可靠的療效預測模式顯得很重要。
    精神分裂症的治療療效是根據一些評估量表分數的變化做評估，例如簡短精神症狀評量表與活性與退化性症狀評量表。本文即比較兩種統計方法建構預測模式的優劣性，第一種是用Multiple Linear Regression直接對量表分數建構預測模式，然後再根據病人的預測分數判斷其是否有達到比治療前分數下降超過20%，依此預測出病人是否有療效。第二種是先判斷若治療後的量表分數比治療前下降20%為有效，否則為無效，然後利用Multiple Logistic Regression建立預測模式。再利用Chang et al [2006]所提出的方法預測出病人的療效。
    我們在幾種不同的情況下比較兩種預測方式的診斷準確度，結果顯示，用線性迴歸預測時會出現將量表分數預測偏低的情況，造成sensitivity高但是specificity很低的結果。而用邏輯式迴歸做出的預測結果，specificity比另外一種方法高但是sensitivity較低。比較兩種預測方法的ROC曲線下面積，邏輯式迴歸預測法都高於線性迴歸預測法

The therapeutic period of schizophrenia needs to last at least for 4 to 8 weeks. Because of the complexity of the disease and the diversity medical responses of patients, it is hard for doctors to evaluate the results of medical response of patients by clinical symptoms and individualities before the treatments. It is intensely important to establish an authentic remedy prediction model in order to avoid the patients to accept ineffective medical treatments.
  The evaluation of the therapeutic effects on schizophrenia is based on the score variation of the certain evaluated scales, for example, Brief Psychiatric Rating Scale and Positive and Negative Syndrome Scale. In the present study, we compared two statistic methods to establish the pros and cons of prediction model. The first statistic method used multiple linear regression to establish prediction model directly by the scores of the scale. The second method determined its effectiveness first. If the scale score after the treatment decreases 20% or more compared with the baseline scale score, the result represents the treatment is effective; otherwise, the treatment is ineffective. Thereafter, we used multiple logistic regression to establish prediction model and then used the method which is brought up by Chang et al. [2006] to predict the effectiveness of the treatment for patients.
  We compared the diagnostic accuracy of two different prediction methods under various circumstances. As the results from the study, when multiple linear regression was used, the predicted scores of the scale tended to be underestimated. The results represented that multiple linear regression has higher sensitivity but lower specificity.  However, the predicted results done by logistic regression has higher specificity but lower sensitivity compare to the results of multiple linear regression.And we compared the areas under the ROC curve of two prediction methods, the area is larger when we use multiple logistic regression to establish a prediction model.

目錄
第一章     前言            1
第二章     文獻回顧        3
第三章     分析方法        5
3.1     準確性指標         5
3.2     線性迴歸預測法     6
3.3     邏輯式迴歸預測法   7
第四章     分析結果比較    8
4.1     收案過程           8
4.2     基本資料分析       9
4.3     分析結果          10
4.3.1   根據第一、二週預測第四、六週    10
4.3.2   根據第一、二、四週預測第六週    13
4.3.3   增加樣本數        15
第五章     討論           17
第六章     結論           20
附錄                      21
參考文獻                  26
表目錄
表1：2×2 Cross Table       5
表2：比較兩種建構預測模式分別對PANSS量表總分及BPRS量
     表總分在第四週與第六週之預測準確性 10
表3：採用預測的CGI改善指標所建構的預測模式下，比較兩種
     預測模式分別對PANSS量表總分及BPRS量表總分在第四
     週與第六週之預測準性  12
表4：在不採用CGI改善指標下建構的預測模式，比較兩種預測
     模式分別對PANSS量表總分及BPRS量表總分在第四週與
     第六週之預測準確性   13
表5：採用CGI改善指標下，比較兩種建構預測模式分別對
     PANSS量表總分及BPRS量表總分在六週之預測準確性 14
表6：採用預測的CGI指標下，比較兩種建構預測模式分別對
     PANSS量表總分及BPRS量表總分在六週之預測準確性 14
表7：不採用預測的CGI指標下，比較兩種建構預測模式對
     PANSS量表總分及BPRS量表總分在六週之預測準確性 15
表8：分別對兩組資料與合併的資料根據BPRS量表總分建構
     預測模式後，在第四週的準確性     15

圖目錄
圖4-2a：BPRS量表分數隨著週期的變化的盒型圖  21
圖4-2b：PANSS量表分數隨著週期變化的盒型圖   21
圖4-2c：第二組資料中BPRS量表分數隨著週期的變化的盒型圖 22
圖4-3-1：PANSS量表與BPRS量表分數，在加入CGI觀測值
         為自變數的清況下，第四週與第六週的ROC curve  23
圖4-3-2：PANSS量表與BPRS量表分數，在加入CGI預測值
         為自變數的清況下，第四週與第六週的ROC curve  24
圖4-3-3：PANSS量表與BPRS量表分數，無加入CGI改善值
         為自變數的清況下，第四週與第六週的ROC curve  25

1.Overall JE, Gorham DR. The Brief Psychiatric Rating Scale. Psychol Rep 1962;10:799-812
2.Kay SR, Fiszbein A, Opler LA. The Positive and Negative Syndrome Scale (PANSS) for schizophrenia. Schizophr Bull 1987;13:261-276
3.Lehman AF, Lieberman JA, Dixon LB, et al. American Psychiatric Association; Steering Committee on Practice Guidelines: Practice guideline for the treatment of patients with schizophrenia, second edition. Am J Psychiatry. 2004;161(suppl 2):1–56.
4.Canadian Psychiatric Association. Canadian clinical practice guidelines for the treatment of schizophrenia. Can J Psychiatry. 1998;43(suppl 2): 25S–40S.
5.Miller AL, Chiles JA, Chiles JK, et al. The Texas Medication Algorithm Project (TMAP) schizophrenia algorithms. J Clin Psychiatry. 1999;60: 649–657.
6.Falkai P, Wobrock T, Lieberman J, et al. WFSBP Task Force on Treatment Guidelines for Schizophrenia: World Federation of Societies of Biological Psychiatry (WFSBP) guidelines for biological treatment of schizophrenia, part 1: acute treatment of schizophrenia. World J Biol Psychiatry. 2005;6:132–191.
7. Falkai P, Wobrock T, Lieberman J, et al. WFSBP Task Force on Treatment Guidelines for Schizophrenia: World Federation of Societies of Biological Psychiatry (WFSBP) guidelines for biological treatment of schizophrenia, part 2: long-term treatment of schizophrenia. World J Biol Psychiatry. 2006;7:5–40.
8.Correll CU, Malhotra AK, Kaushik S, et al. Early prediction of antipsychotic response in schizophrenia. Am J Psychiatry. 2003;160:2063–2065.
9.Yue-Cune Chang, PhD, Hsien-Yuan Lane, MD, PhD, et al. Optimizing Early Prediction for Antipsychotic Response in Schizophrenia. J Clin. Psychopharmacol 2006;26:554-559.
10.Lusted LB. Decision-making Studies in Patient management. N Engl. J Med 1971, 284: 416-423
11.Metz CE. Basic Principles of ROC analysis. Sem Nucl Med 1978, 8: 283-298
12.American Psychiatric Association. Structured Clinical Interview for DSM-IV. Washington, DC, American Psychiatric Press. 1994.