Propensity Scores and Preventive Drug Use in the Elderly

说明(申请人提供)：预防性使用处方药和非处方药是照顾老年人的一个主要特点。评估老年人预防性药物有效性的随机对照试验费用高昂，而且由于参与者的选择、治疗持续时间较短以及相互竞争的结果，其推广可能受到限制。然而，对这些问题的观察性研究经常因结果有偏见而受到批评。因此，在评估老年患者用药预防效果的观察性研究中，开发和应用适当的方法来减少偏差是至关重要的。在这类药物观察性研究中，倾向评分和多变量混杂评分被提出用来改善混杂的控制。尽管使用量在增加，但几乎没有证据表明混杂得到了更好的控制。 我们建议与传统的多变量分析相比，评估倾向评分和相关技术在研究老年人用药效果方面的有效性、统计效率和适用性。 预防性使用非甾体类抗炎药与结直肠癌之间的关系作为我们的主要例子来阐明倾向评分和多变量混杂评分如何能够减少偏倚。我们将使用两项大型队列研究--医生健康研究和妇女健康研究--来评估这种联系。由于这项研究侧重于评估此类数据的分析方法，而不是对任何单一生物机制的评估，因此我们探索了预防性药物使用和健康结果之间的其他可能的关联，这些关联是故意选择的，以反映广泛的典型分析情况，作为我们多变量模拟研究的基础。 基于这些大型队列研究的结果，我们将进行广泛的统计模拟，以评估一系列现实情景下倾向分数和多变量混杂分数的有效性和效率提高的决定因素。 结果将被整合到有意义和易于使用的决策规则和实用建议中，以便在老年人的观察研究中适当使用倾向分数和多变量混杂分数。这些建议将在研究人员和执业医生中广泛传播。

DESCRIPTION (provided by applicant): The preventive use of prescription- and non-prescription drugs is a central feature of the care of the elderly. Randomized controlled trials assessing the efficacy of preventive medications in the elderly are costly and may be limited in their generalizability due to selection of participants, shorter duration of treatment, and competing outcomes. However, observational studies of these questions are frequently criticized for biased results. It is therefore vital to develop and apply adequate methods to reduce bias in observational studies that assess the preventive effects of medication use in older patients. Propensity scores and multivariate confounder scores have been proposed to improve control for confounding in such observational studies of medications. Despite increasing use, there is little evidence that better control of confounding is achieved. We propose to assess the validity, statistical efficiency, and applicability of propensity scores and related techniques compared to 'traditional' multivariate analysis in studying the effects of medications in the elderly. The relation between preventive NSAID use and colorectal cancer serves as our primary example to elucidate how propensity scores and multivariate confounder scores can reduce bias. We will use two large cohort studies the Physicians' Health Study and the Women's Health Study to evaluate this association. Because the study focuses on the analytic approach for evaluating such data, and not on the assessment of any single biological mechanism, we explore other possible associations between preventive drug use and health outcomes that were deliberately selected to reflect a wide range of typical analytic situations as the basis for our multivariate simulation studies. Based on the results of these large cohort studies we will conduct extensive statistical simulations to assess the determinants of improved validity and efficiency of propensity scores and multivariate confounder scores for a range of realistic scenarios. Results will be integrated into meaningful and easy-to-use decision rules and practical recommendations for the appropriate use of propensity scores and multivariable confounder scores in observational research in the elderly. These recommendations will be disseminated widely among researchers and practicing physicians.