| 摘要: | 精神分裂症是一種錯綜複雜且頗具異質性的精神疾病，其藥物治療的臨床試驗療程通常至少持續42天，並在治療前與治療後的第14、28和42天進行療效評估。因此，資料結構形成了所謂的長期重複測量的相依資料。本研究將近年來陸續發表的20篇文章中所探討的三大類可能影響療效之危險因子重新彙整，以PANSS及多個評估量表分數，如CGI表，作為指標(依變項)，利用處理相依資料廣義線性模式之GEE方法和一些探討相關性的無母數方法，加以釐清上述三大類危險因子中的調節效應(moderation)與仲介效應 (mediation)。由於精神分裂症療效評估之相關研究，主要是探討各項臨床診斷指標隨著時間改變情形。因此，在探討調節效應方面，我們將重點放在釐清那些因子會影響時間因子對各指標關係的調節效應，亦即扮演調節因子(moderator)的角色。依此觀念，在探討仲介因子(mediator)時，必須先行調整時間效應，再利用GEE方法並參酌各變項的特性，必要時選用Chi-Sqiare Test、無母數方法或是Kendall’s bivariate correlations來確定各危險因子對各項指標的仲介效應 (mediation)。由結果來看，治療之時間效應對九個療效指標的關係都會受到遺傳基因纇的危險因子的影響；疾病相關類的危險因子則會影響治療時間對除了CGI外的八個療效指標的關係；病患個人特質類的危險因子對治療時間與各項療效指標間的關係的影響則較少，僅影響了時間對CGI、PANSS負性症狀量表、PANSS總分與其對應之療效指標(response，定義為20% reduction)等四個指標的關係。至於進一步探討之部份仲介因子(partial mediator)僅有遺傳基因與類疾病相關類的危險因素這兩類，而完全仲介因子(complete mediator)方面，只有體重為性別對PANSS負性症狀得分的完全仲介因素。
Schizophrenia is a complex and heterogeneous psychosis. The durations of antipsychotic trial for drug treatment of schizophrenia recommend to at least 42 days. Efficacy and drug safety were assessed on days 0, 14, 28, and 42. Therefore, the data structure is a longitudinal dependent data. We have published 20 papers to explore the relationship between the treatment effects of schizophrenia and some potential prognostic/risk factors. Those factors can be classified into three major areas: genes, disease related factors and patients’ personal characteristics. Efficacy was measured by the PANSS and/or other clinical evaluation forms like Clinical Global Impression (CGI), named dependent variables. We used GEE method’s generalized linear models/other non-parametric methods to explore the possible moderation effects and/ the mediation effects among those aforementioned risk factors. One of the primary study purposes of evaluating the treatment effects of schizophrenia is to figure out the changes of each response variable with respect to time. Therefore, to explore the moderation effect, we focus on the moderation effect on the treatment time to each response variable. Accordingly, to explore the mediation effect, we used the GEE method’s generalized linear model to adjust the time effect. Some other nonparametric methods, such as Chi-Square Test, Kruskal-Wallis Test, Kolmogorov-Smirnov Test, or Kendall’s bivariates correlations, were used to explore the mediation effect of those risk factors to each response variable according to the characteristics of risk factors. In conclusion, the relationships between treatment week and those nine response variables were influenced by the genetic factors; the disease related factors influence the relationships between treatment week and the response variables (except CGI); the patients’ characteristic factors were influence on the relationships between treatment week and four indicators ( CGI, PANSS negative score, PANSS total score, and response ). Further analyses on the partial/complete mediator, we found that some genetic and disease related factors were partial mediators. And, body weight was the only complete mediator of sex to PANSS negative score. |
