Covariance matrix estimation in matched case-control studies
Conference
64th ISI World Statistics Congress - Ottawa, Canada
Format: IPS Abstract
Keywords: covariance, epidemiology, high-dimensional
Abstract
Covariance matrix estimation is an essential part of multivariate analysis since it represents the marginal dependence structures between the variables. Here, we investigate covariance matrix estimation when data are sampled under matched case control study designs. The matched case-control study design is a common and efficient way of studying rare diseases or illnesses with long latency periods. Matching of cases to controls is frequently employed to control the effects of known potential confounding variables, such as age, biological sex and others. In a matched case-control study, a case is usually matched with one or more controls on potential confounders. Due to the nature of sampling, however, matched datasets are not random samples from the target population.
By incorporating inverse probability of sampling weights, we proposed a weighted covariance estimator appropriate when p > n). Through simulation studies, we show that our proposed methods outperform direct estimation that ignores the matched sampling. We illustrate the methods proposed in a biomedical investigation in which matched sampling was employed.