傳統精神分裂症的藥物治療之療效評估至少需費時六週。為此，臨床醫師們多年來努力追求於精神分裂症的治療療效能被早期預測出來。而早期研究成果中，用來預測療效的邏輯斯迴歸模式所用的獨立變數中有一項CGI，其與診斷有效與否的PANSS同為評估療效的臨床診斷表，意即兩者同為Outcome Variable。若將CGI當為獨立變數使用，在臨床應用上頗不合理。有鍳於此，本論文採用兩階段方法建構預測模型。第一階段先針對CGI之得分(或改善與否)做預測，再將其預測結果當做第二階段預測PANSS總分(或有效與否)的獨立變數。上述各種預測方法的預測準確性之比較，我們以敏感度、特異度、陽性預測值(PPV)、陰性預測值(NPV)及預測力(PP)做為預測模型優劣評判的指標。 The therapeutic period of schizophrenia needs to last ad least for 6 weeks. For this reason, clinician pursue treatment curative effect in schizophrenia can predict, come out in early days hard for many years. And in the early research results, the logit used for predicting curative effect returns to the independent variable that the way uses then and has a CGI, it, with diagnosing PANSS effective is the clinical diagnosis form which assesses curative effect together, mean the two are all Outcome Variable. If regard CGI as and use for the independent variable, quite unreasonable in clinical practice. This thesis adopts two stage methods to build and construct the model for predicting. First stage score (or improving or not) to CGI first Do, predict, predict it result regard as, predict PANSS total points (or effective or not) two stage and then Independent variable. The comparison of prediction accuracy of different prediction methods described above, we are with sensitivity, specificity, positive predicted value (PPV) , negative predicted value (NPV) and predict power (PP). As predicting the good and bad indicator that judges of model.