Statins and Risk of Lower Limb Revision Surgery: The Influence of Differences in Study Design Using Electronic Health Records From the United Kingdom and Denmark
Statins and Risk of Lower Limb Revision Surgery: The Influence of Differences in Study Design Using Electronic Health Records From the United Kingdom and Denmark
复制标题
DOI：
10.1093/aje/kwv311
复制
发表时间：
2016-07-01
影响因子：
5
通讯作者：
de Vries, Frank
中科院分区：
文献类型：
--
作者：
Lalmohamed, Arief;van Staa, Tjeerd P.;de Vries, Frank
先前关于他汀类药物的观察性研究显示了基于所用方法的可变结果。我们的目的是研究他汀类药物与骨科植入物失败之间的关系，并探讨研究设计中方法学差异的影响。我们的研究基础包括在丹麦和英国接受初次全关节置换术的患者（n = 189，286; 1987-2012）。我们使用了4种研究设计：1）病例对照（每例翻修手术患者与4例对照匹配），2）时间依赖性队列（术后他汀类药物使用作为随时间变化的暴露变量），3）不朽时间队列（对术后他汀类药物使用前的时间进行错误分类），4）时间排除队列（排除术后他汀类药物使用前的时间）。采用考克斯比例风险模型和逻辑回归估计发病率比。在时间依赖性队列设计中，他汀类药物的使用与翻修手术风险降低相关（校正发生率比（IRR）= 0.90，95%置信区间（CI）：0.85，0.96），这与我们的病例对照结果相似（IRR = 0.87，95% CI：0.81，0.93）。相比之下，两种固定时间队列设计的风险估计值均大幅降低（IRR = 0.36（95%CI：0.34，0.38）和IRR = 0.65（95%CI：0.63，0.68））。我们不鼓励使用固定时间的队列研究，因为这可能会错误地暗示保护作用。如何对暴露进行分类的简单选择可以大大改变结果，从生物学上合理到难以置信。
Previous observational studies on statins have shown variable results based on the methodology used. Our objective was to study the association between statins and orthopedic implant failure and to explore the influence of methodological differences in study design. Our study base consisted of patients with a primary total joint replacement in Denmark and the United Kingdom (n = 189,286; 1987-2012). We used 4 study designs: 1) case-control (each patient with revision surgery matched to 4 controls), 2) time-dependent cohort (postoperative statin use as a time-varying exposure variable), 3) immortal time cohort (misclassifying the time postoperatively before statin use), and 4) time-exclusion cohort (excluding the time postoperatively before statin use). Cox proportional hazards models and logistic regression were used to estimate incidence rate ratios. In the time-dependent cohort design, statin use was associated with a decreased risk of revision surgery (adjusted incidence rate ratio (IRR) = 0.90, 95% confidence interval (CI): 0.85, 0.96), which was similar to our case-control results (IRR = 0.87, 95% CI: 0.81, 0.93). In contrast, both time-fixed cohort designs yielded substantially lower risk estimates (IRR = 0.36 (95% CI: 0.34, 0.38) and IRR = 0.65 (95% CI: 0.63, 0.68), respectively). We discourage the use of time-fixed cohort studies, which may falsely suggest protective effects. The simple choice of how to classify exposure can substantially change results from biologically plausible to implausible.