Byar D, Chastang C
Bull Cancer. 1987;74(2):177-84.
For most diseases, one observes great variability in clinical course and response to therapy which should be taken into account in the design and the analysis of clinical trials. Eligibility criteria and stratified randomization use prognostic knowledge. A suitable analysis of randomized studies includes: (1) checking treatment effect groups for comparability, (2) adjustment of treatment effect by stratifying or modeling to account for any imbalances in prognostic factors and to increase the precision of the treatment comparison by reducing random variation, (3) looking for treatment-covariate interactions suggesting that response to treatment depends on patients' characteristics. Tests for interaction should be interpreted with caution in light of medical plausibility; multiple comparisons should be taken into account. Recording information on known prognostic factors in clinical trials is important, as is the continuing search for new ones.
对于大多数疾病，人们观察到临床病程和对治疗的反应存在很大差异，在临床试验的设计和分析中应考虑到这一点。入选标准和分层随机化运用了预后知识。对随机研究进行适当分析包括：（1）检查治疗效果组的可比性；（2）通过分层或建模调整治疗效果，以考虑预后因素的任何不平衡，并通过减少随机变异提高治疗比较的精度；（3）寻找治疗-协变量相互作用，表明对治疗的反应取决于患者特征。鉴于医学合理性，对相互作用的检验应谨慎解释；应考虑多重比较。在临床试验中记录已知预后因素的信息很重要，持续寻找新的预后因素也同样重要。
