Linden Ariel, Adams John L
Oregon Health & Science University, School of Medicine, Portland, OR, USA.
J Eval Clin Pract. 2008 Oct;14(5):914-8. doi: 10.1111/j.1365-2753.2008.01091.x.
While the randomized controlled trial (RCT) remains the gold-standard study design for evaluating treatment effect, outcomes researchers turn to powerful quasi-experimental designs when only observational studies can be conducted. Within these designs, propensity score matching is one of the most popular to evaluate disease management (DM) programme effectiveness. Given that DM programmes generally have a much smaller number of participants than non-participants in the population, propensity score matching will typically result in all or nearly all participants finding successful matches, while most of the non-participants in the population remain unmatched and thereby excluded from the analysis. By excluding data from the unmatched population, the effect of non-treatment in the remaining population with the disease is not captured. In the present study, we examine changes in hospitalization rates stratified by propensity score quintiles across the entire population allowing us to gain insight as to how well the programme chose its participants, or if the programme could have been effective on those individuals not explicitly targeted for the intervention. These data indicate the presence of regression to the mean, and suggest that the DM programme may be overly limited to only the highest strata when there is evidence of a potential benefit for those in all the lower strata as well.
虽然随机对照试验（RCT）仍然是评估治疗效果的金标准研究设计，但当只能进行观察性研究时，结果研究人员会转向强大的准实验设计。在这些设计中，倾向得分匹配是评估疾病管理（DM）项目有效性最常用的方法之一。鉴于DM项目的参与者在总体中通常比非参与者少得多，倾向得分匹配通常会导致所有或几乎所有参与者找到成功匹配，而总体中的大多数非参与者仍然无法匹配，从而被排除在分析之外。通过排除未匹配人群的数据，未治疗对其余患病人群的影响就无法体现。在本研究中，我们检查了整个人口中按倾向得分五分位数分层的住院率变化，这使我们能够深入了解该项目如何选择参与者，或者该项目对那些未被明确列为干预对象的个体是否有效。这些数据表明存在均值回归现象，并且表明当有证据显示较低分层中的个体也可能受益时，DM项目可能过度局限于最高分层。
