G Ronco1, A Biggeri. 1. Unit of Cancer Epidemiology, Department of Oncology, Torino, Italy.
Abstract
BACKGROUND: When diagnostic tests are repeated on the same subject, a correlation between test results is expected. This must be taken into account when computing interval estimates. METHODS: The application of regression techniques for correlated data to a logistic model is proposed in order to obtain correct point and interval estimates. RESULTS: The use of different study designs and models of analysis (marginal and 'random effects') is discussed. The advantages of studies using multiple tests, possible sources of bias (including those specific to multiple testing) and methods of correction are also discussed. An example considering Doppler flow velocimetry in pregnancy is presented.
BACKGROUND: When diagnostic tests are repeated on the same subject, a correlation between test results is expected. This must be taken into account when computing interval estimates. METHODS: The application of regression techniques for correlated data to a logistic model is proposed in order to obtain correct point and interval estimates. RESULTS: The use of different study designs and models of analysis (marginal and 'random effects') is discussed. The advantages of studies using multiple tests, possible sources of bias (including those specific to multiple testing) and methods of correction are also discussed. An example considering Doppler flow velocimetry in pregnancy is presented.
Authors: Binhua Ling; Mario L Santiago; Sreelatha Meleth; Bobby Gormus; Harold M McClure; Cristian Apetrei; Beatrice H Hahn; Preston A Marx Journal: J Virol Date: 2003-02 Impact factor: 5.103