Methodological progress note: A clinician's guide to propensity scores
Methodological progress note: A clinician's guide to propensity scores
复制标题
DOI：
10.1002/jhm.12791
复制
发表时间：
2022-02-14
影响因子：
2.6
通讯作者：
Hall, Matthew
中科院分区：
文献类型：
--
作者：
Ambroggio, Lilliam;Cotter, Jillian M.;Hall, Matthew
虽然随机对照试验 (RCT) 被许多人认为是临床研究的黄金标准，但随机化并不总是可能的（例如，伦理问题或高昂的成本）。在这些情况下，一个可行的替代方案是使用观察数据并应用适当的统计方法来模拟随机研究。与随机对照试验中随机分配给患者治疗不同，在观察性研究中，我们“观察”每个患者接受的治疗作为临床护理的一部分。然而，观察结果可能会受到治疗以外的因素的影响，因此需要控制其影响。利用此类数据，倾向方法可以估计治疗对结果的影响。 1 例如，在一项观察性研究中，一名儿童可能接受阿莫西林治疗肺炎，而另一名儿童可能接受头孢曲松治疗。医生偏好、药物过敏、儿童年龄和疾病严重程度等因素都会影响使用一种或另一种药物治疗的决定。为了解释可能与治疗决策相关的患者组差异（例如，导致选择头孢曲松的疾病严重程度），在评估治疗（例如，阿莫西林或头孢曲松）与目标结果（例如，住院时间）之间的关联时，需要考虑所有这些因素。倾向方法是通过两步过程实现这一目标的一种方法：(a) 制定倾向评分，以及 (b) 使用所得倾向评分来估计治疗对结果的影响。倾向方法可用于评估任何治疗、设备、行为或其他干预措施。为简单起见，我们将一般使用术语“干预”来指代可以分配给个体患者的任何事情。在本进度说明中，我们将讨论何时使用倾向方法、如何创建和使用倾向得分以及如何解释结果。
While randomized controlled trials (RCTs) are considered by many to be the gold standard of clinical research, randomization is not always possible (eg, ethical concerns or prohibitive costs). In these instances, a viable alternative is to mimic a randomized study using observational data and applying the appropriate statistical methodology. Unlike in RCTs where treatments are randomly assigned to patients, in observational studies we “observe” which treatment each patient received as part of their clinical care. However, observations can be influenced by factors other than the treatment received, creating a need to control for their effects. With this type of data, propensity methods can estimate the effect of a treatment on the outcome. 1 For instance, in an observational study, a child may receive amoxicillin to treat pneumonia, whereas another child may receive ceftriaxone. Factors such as physician preference, drug allergy, age of the child, and severity of the illness can all influence the decision to treat with one or the other drug. To account for differences in the patient groups that may be related to the treatment decision (eg, severity of illness driving the choice of ceftriaxone), all these factors need to be accounted for when assessing the association between the treatment (eg, amoxicillin or ceftriaxone) and the outcome of interest (eg, length of stay). Propensity methods are one way of accomplishing this through a two-step process:(a) development of the propensity score, and (b) using the resulting propensity scores to estimate the effect of the treatment on the outcome.Propensity methods can be applied to evaluate any treatment, device, behavioral or other intervention. For simplicity, we will use the term “intervention” generically to refer to anything that can be assigned to an individual patient. In this progress note, we will discuss when to use propensity methods, how to create and use the propensity score, and how to interpret the results.