随机对照试验中简单随机和最小化病例分配方法不可预测性的对比研究

Comparison of the unpredictability between the simple randomized allocation method and the minimized case method in randomized controlled trials

目的比较病例分配方法中完全随机和最小化方法分配结果的不可预测性。方法采用340例数据的真实随机对照试验数据,使用两种分配方法分配后,分别使用Logistics回归、随机森林和线性核支持向量机对已分配病例建模并预测下个病例的分配结果。以正确预测比例对比两种方法的不可预测性。结果完全随机方法达到几乎完美的不可预测性,最小化方法有一定可预测性。结论对于随机对照试验,应在保证组间平衡的前提下,尽量减小分配偏倚因子。

Objective To compare the unpredictability of the allocation results of a completely randomized method and the minimized method in case allocation methods. Methods The real randomized controlled trial data containing 340 cases were used and allocated with the above two allocation methods. Logistics regression, random forest and linear kernel support vector machine were applied respectively to predict the next case. The unpredictability of the two allocation methods was compared with the correct predictor ratio. Results The results of completely randomized method achieved almost perfect unpredictability, and the minimization method had some predictability. Conclusion For randomized controlled trials, the allocation bias factor should be minimized under the premise of ensuring a balance between groups.