Objective To develop a new statistical method for comparing effectiveness of two treatments for the same disease indication. A statistical method needs to be developed to compare effectiveness of different treatments by using existing clinical data when lacking of evidence from randomized controlled trials. Methods Datasets including an outcome variable and a confounder for two treatments were simulated. Subjects in each dataset were randomly divided into two arms for 100 times, and the means were compared by using student-ttests and Per-protocol analysis strategy was applied. Subjects with assigned treatment by randomization that was contradictory to the actual received treatment were considered drop-out of conventional RCTs, and were excluded from comparisons of means. The ratio of the frequency of hypothesis tests that rejected H0 to the frequency of hypothesis tests that did not reject H0 , called odds, was used to predict the probability of significant tests for group difference. To document the consistency and reliability of the method, the distribution of odds and its 95% CI, and the confounding effect were described in simulated datasets with various between-group mean differences and statistical power (ranging from 0. 5 to 0. 85) with varied sample sizes (n =50, 100, 500, and 1 000). Stata 11.0 was used to program and perform the analysis. Results The odds and its 95% CI of simulated RCTs were perfectly and linearly correlated with the change of between-group mean differences and statistical power, by differencesample size. A conclusion could be made based on the hypothesis of the simulated randomized controlled trials ( sRCT). The probability of loss of balance of confounding was over 95% for equal and unequal sample size of two arms after excluding misclassified subjects. Conclusions The proposed novel analytical method, simulated RCTs based on real clinical treatment data, can be used to compare effectiveness of two treatments when evidence from RCTs is unavailable.%目的 本研究旨在通过探索处理混杂因素的手段,创造一种新的用于观察性数据疗效比较研究的统计分析方法.方法 本方法基于以下原理:针对诊断为同种疾病接受不同治疗的一组患者,采用反复多次模拟随机化分组并根据RCT的统计分析策略进行疗效比较,以拒绝H0的试验频率和不拒绝H0的试验频率之比(odds值)及其95％CI作为判断不同治疗方法间疗效差异的依据.采用计算机模拟的方法获得统计量odds值的分布.对包含结局变量和混杂因素变量的模拟数据库进行随机化分组,对根据符合方案集分析(PP)策略保留下来的样本进行结局变量比较.重复100次随机化分组,并对每次随机化分组后结局变量进行比较,同时也对混杂因素变量的组间均衡性进行分析.计算100次结局变量比较分析结果中拒绝H0与不拒绝H0的比值,即odds值,重复100次odds值的计算过程得到odds值的点估计值及其95％CI.根据样本量(n1=n2 =50,100,500和1 000)、组间差异的把握度和效应量产生多个模拟数据库,观察分析得到的odds值及其95％CI的一致性和稳定性.同时验证混杂因素在根据PP策略保留下来的样本的组间均衡性.结果 ①对不同样本量下疗效有差异数据库分析得到的odds值均＞1,odds值及其95％ CI均随把握度的增加呈上升趋势；②对不同样本量下疗效无差异数据库分析得到的odds值均＜1,odds值及其95％CI均随把握度的增加呈下降趋势,二者变化均呈现良好的线性关系；③同时验证样本量相等和不相等的情况下,混杂因素组间均衡的概率均＞95％.结论 将本文发明的方法命名为模拟随机对照试验方法,简称sRCT.运用sRCT对模拟数据库分析得到结果的一致性和稳定性高,实现了在均衡混杂因素的基础上,创建了一种用于观察性数据疗效比较研究的新方法.
展开▼
