Propensity score matching for estimation of pairwise marginal hazard ratios

There is a growing interest in using observational studies to estimate the effects of treatments on survival or time-to-event outcomes. However, few standard approaches can adequately accommodate multiple treatment levels, which are common in observational comparative effectiveness research. We study the asymptotic properties of the generalized propensity score matching estimators of the marginal hazard ratios between pairs of treatment levels. The estimates are obtained by fitting a marginal Cox proportional hazard model on the matched dataset. We evaluate our approach in a simulation study and a case study where we analyze the IQVIA electronic medical records data.

当前，利用观察性研究评估治疗手段对生存结局或事件发生时间结局的影响这一方向正受到日益广泛的关注。然而，现有标准方法鲜有能够妥善适配多治疗水平场景的——而这类场景在观察性比较有效性研究中极为常见。本文研究了治疗水平两两间边际风险比的广义倾向得分匹配估计量（generalized propensity score matching estimators）的渐近性质，该类估计通过在匹配后的数据集上拟合边际Cox比例风险模型得到。本文通过一项模拟研究与一项案例研究对所提方法进行了评估，其中案例研究分析了IQVIA电子医疗记录数据集。