Comparison of nested case-control and survival analysis methodologies for analysis of time-dependent exposure.
Comparison of nested case-control and survival analysis methodologies for analysis of time-dependent exposure.
复制标题
DOI：
10.1186/1471-2288-5-5
复制
发表时间：
2005-01-25
影响因子：
4
通讯作者：
Pilote, Louise
中科院分区：
文献类型：
--
作者：
Essebag, Vidal;Platt, Robert W;Pilote, Louise
背景：随时间变化的暴露的流行病学研究需要额外的方法复杂性来解释暴露的时间依赖性。本研究比较了时间依赖性暴露研究的巢式病例对照方法与使用含时间依赖性协变量的Cox回归的队列分析。方法：采用包含4个固定协变量和7个时间相关协变量的1340名受试者的队列进行研究。对每个病例的4、8、16、32和64个对照重复进行巢式病例-对照分析100次，并将点估计与在全队列中使用Cox回归获得的结果进行比较。通过比较在队列大小为初始大小的1、2、4、8、16和32倍时分析所需的中央处理单元时间来评估计算效率。结果：嵌套病例对照分析的结果与全队列Cox回归的结果相似。发现Cox回归比嵌套病例-对照方法慢125倍（每个病例使用4个对照）。结论：在研究时间依赖性暴露时，嵌套病例对照方法是一种有用的替代队列分析方法。在研究数据库中的罕见结果时，其优越的计算效率可能特别有用，在这种情况下，分析更大样本量的能力可以提高研究的能力。
BACKGROUND: Epidemiological studies of exposures that vary with time require an additional level of methodological complexity to account for the time-dependence of exposure. This study compares a nested case-control approach for the study of time-dependent exposure with cohort analysis using Cox regression including time-dependent covariates.METHODS: A cohort of 1340 subjects with four fixed and seven time-dependent covariates was used for this study. Nested case-control analyses were repeated 100 times for each of 4, 8, 16, 32, and 64 controls per case, and point estimates were compared to those obtained using Cox regression on the full cohort. Computational efficiencies were evaluated by comparing central processing unit times required for analysis of the cohort at sizes 1, 2, 4, 8, 16, and 32 times its initial size.RESULTS: Nested case-control analyses yielded results that were similar to results of Cox regression on the full cohort. Cox regression was found to be 125 times slower than the nested case-control approach (using four controls per case).CONCLUSIONS: The nested case-control approach is a useful alternative for cohort analysis when studying time-dependent exposures. Its superior computational efficiency may be particularly useful when studying rare outcomes in databases, where the ability to analyze larger sample sizes can improve the power of the study.